SlideShare a Scribd company logo
1 of 14
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
 IEEE TRANSACTIONS ON MOBILE COMPUTING

                                                                                                                                                                         1




               Estimating Parameters of Multiple
         Heterogeneous Target Objects Using Composite
                        Sensor Nodes
          Hiroshi Saito, Fellow, IEEE, Shinsuke Shimogawa, Sadaharu Tanaka, and Shigeo Shioda, Member, IEEE

                 Abstract—We propose a method for estimating parameters of multiple target objects by using networked binary sensors whose
                 locations are unknown. These target objects may have different parameters, such as size and perimeter length. Each sensors, which is
                 incapable of monitoring the target object’s parameters, sends only binary data describing whether or not it detects target objects coming
                 into, moving around, or leaving the sensing area at every moment. We previously developed a parameter estimation method for a single
                 target object. However, a straight-forward extension of this method is not applicable for estimating multiple heterogeneous target objects.
                 This is because a networked binary sensor at an unknown location cannot provide information that distinguishes individual target
                 objects, but it can provide information on the total perimeter length and size of multiple target objects. Therefore, we propose composite
                 sensor nodes with multiple sensors in a predetermined layout for obtaining additional information for estimating the parameter of each
                 target object. As an example of a composite sensor node, we consider a two-sensor composite sensor node, which consists of two
                 sensors, one at each of the two end points of a line segment of known length. For the two-sensor composite sensor node, measures
                 are derived such as the two sensors detecting target objects. These derived measures are the basis for identifying the shape of each
                 target object among a given set of categories (for example, disks and rectangles) and estimating parameters such as the radius and
                 lengths of two sides of each target object. Numerical examples demonstrate that networked composite sensor nodes consisting of two
                 binary sensors enable us to estimate the parameters of target objects.

                 Index Terms—Sensor network, estimation, target object, composite sensor node, ubiquitous network, integral geometry, geometric
                 probability.

                                                                                        ✦



         1      I NTRODUCTION                                                               extract significant information. This sensing paradigm is
                                                                                            appropriate for the described situation.
         Small and low-cost sensors can be used to build wireless
                                                                                              Unfortunately, we have not determined a theory that
         sensor networks [1], [2], [3]. They have communication
                                                                                            enables us to implement this paradigm. As a typical
         capabilities and built-in batteries to communicate via a
                                                                                            and challenging application, we investigated a problem
         wireless link for transmitting sensed data of detected
                                                                                            of estimating parameters related to the shape and size
         events of interest. A newly proposed development, such
                                                                                            of target objects moving in a monitored area by using
         as a wide area ubiquitous network [4], supports sensors
                                                                                            many simple randomly distributed sensors. We address
         with low performance and functionalities and a long-
                                                                                            this problem and present a theory for estimating the
         range, low-speed wireless link with very low power con-
                                                                                            parameters of multiple target objects by using networked
         sumption. These sensors can be implemented as single
                                                                                            binary sensors whose locations are unknown. An in-
         chips, lowering their cost. As a result, we can deploy
                                                                                            dividual sensor is simple. It monitors its environment
         such sensors like the scattering of dust [5]. Because there
                                                                                            and reports whether or not it detects a target object. It
         are many sensors, we cannot carefully design the loca-
                                                                                            does not have a positioning function, such as a GPS, or
         tion of each one. GPS cannot be used because the sensors
                                                                                            functions for monitoring the target object size and shape,
         should have limited capability and power consumption.
                                                                                            such as a camera, and it can be placed without careful
            This situation leads us to a new sensing paradigm.
                                                                                            design. However, by collecting reports from individual
         That is, instead of having a few sensors with advanced
                                                                                            sensors, we can statistically estimate the parameters of
         functions and high performance, many sensors with
                                                                                            target objects.
         simple functions and low performance are distributed
                                                                                              In addition to theoretical contributions, this work is
         randomly. They are networked and send reports, each of
                                                                                            useful to some application scenarios that require low
         which includes only an insignificant amount of informa-
                                                                                            cost, a wide coverage, and low power consumption
         tion, over a wireless link. However, as a whole, we can
                                                                                            and accept rough estimation of the size and shape of
                                                                                            target objects. For example, we may need to identify the
         ¯ H. Saito and S. Shimogawa are with NTT Service Integration Laboratories,
           3-9-11, Midori-cho, Musashino-shi, Tokyo, 180-8585, Japan.
                                                                                            number and kinds of animals and vehicles in a large
           E-mail: saito.hiroshi@lab.ntt.co.jp                                              area to observe the wildlife, prevent poaching, or limit
         ¯ S. Tanaka and S. Shioda are with Chiba University.                               the number of tourist vehicles. We can easily obtain bi-
                                                                                            nary sensors of sound volume, acceleration, infrared, or



Digital Object Indentifier 10.1109/TMC.2011.65                         1536-1233/11/$26.00 © 2011 IEEE
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING

                                                                                                                                                                        2



       magnetic sensors for such scenarios. These sensors with                            for such a purpose. In particular, [8] and [14] directly
       normal detection ranges are not expensive. Therefore, we                           apply the results of integral geometry discussed in
       can provide a large number of these sensors to cover a                             Chapter 5 and Section 6.7 in [15] to the analysis of
       large area at a reasonable cost. By distributing these bi-                         detecting an object moving in a straight line and to the
       nary sensors without expensive power-consuming GPSs                                evaluation of the probability of -coverage. In addition,
       or time-consuming carefully-designed placement in a                                [16] applies integral geometry to the analysis of straight
       large area, our proposed method can work.                                          line routing, which is an approximation of the shortest
          This research is an extension of our prior study [6],                           path routing, and [17] uses it to operate sensors in an
       which was based on the coverage process theory [7]                                 energy-conserving way.
       and its application to sensor networks [8], [9], [10]. This                           The rest of this paper is organized as follows.
       study emphasized the robustness of the derived formula                                Section 2 describes a basic model and previous results
       for a number of sensors detecting a target object. In                              for the model. In Section 3, we introduce composite
       [6], sensors that measure the size of a detected part                              sensor nodes and describe an extended model (static
       of a target object are used for estimating the overall                             submodel). We discuss the analysis for deriving mea-
       size. However, in [11], we developed a shape and size                              sures in Section 4. In Section 5, we discuss the dynamic
       estimation method using binary sensors for sending                                 submodel of the extended submodel. We then propose
       reports on whether or not they detect the target object.                           an estimation method that uses composite sensor nodes
       That study assumed that both the target object and the                             for target objects with different parameters in Section 6.
       sensing area are convex or that the sensing area is disk-                          We show numerical examples in Section 7, and conclude
       shaped. This developed method [11] was evaluated in                                this paper in Section 8.
       an experiment where the target object was a box and
       the sensor was an infrared distance measurement sensor
       [12]. We also extended the estimation method in [11] to                            2     P REVIOUS          WORK
       apply to non-convex target objects [13].
                                                                                          2.1    Basic model
          These methods we developed in [11], [13] did not
       take into account when multiple target objects are in the                          To describe the results of the previous work, which is
       monitored area. This work addresses such a case, i.e.,                             the basis of the current work, we discuss the basic model
       multiple target objects that may have different shapes                             that the previous work used.
       and sizes coming into, moving around, and leaving the                                 A sensor network operator deploys sensors in a con-
       area monitored by sensors. Each sensor, whose location                             vex area ¨ in 2-dimensional space ʾ , but the operator
       is unknown, sends a report on whether or not at least                              does not know the sensor locations. Each sensor has
       one target object is detected within its sensing area. The                         a sensing area, monitors the environment, and detects
       sensor network operator collects these reports to identify                         events within that area. (This model is called the Boolean
       the shape and to estimate parameters, such as sizes                                sensing model [9], [18], [19], [20] because it is clear
       and perimeter lengths, of the target objects. However,                             whether a point is within a sensing area or not.) A sensor
       identifying the shape and estimating the parameters of                             sends a detection report if a target object is in its sensing
       multiple heterogeneous (i.e., having different parame-                             area, and it sends a no-detection report if there is no
       ters) target objects is difficult, and a straight-forward                           target object in that area. (It is possible to assume that a
       extension of the estimation method we previously de-                               sensor does not send a report if there is no target object
       veloped [11] is not applicable. Thus, we introduce the                             in that area and that the sensor network operator judges
       concept of composite sensor nodes where multiple sim-                              no-detection if there is no report within a timeout. This
       ple binary sensors are arranged in a predetermined                                 scenario can reduce the power consumption of sensors,
       layout, such as a line. Sensors in a composite sensor                              but the sensor network operator cannot distinguish no-
       node provide local and relative information even if the                            detection from the loss of reports. For simplicity, this
       location of that node is unknown. This is because the                              paper assumes that no-report is sent if there is no target
       relative locations of these sensors are predetermined. In                          object in that area.)
       this sense, the composite sensor node is an intermediate                              Assume that the -th sensor is located at ´Ü Ý µ and
       concept between a simple and an advanced sensor node                               the sensing area is rotated by          from the referenced
       equipped with GPS. With these sensors, we can estimate                             position. Let ´Ü Ý         µ       ʾ be the sensing area
       the parameters of multiple heterogeneous target objects.                           where        ½ ¾ . The -th sensor has communication
          The estimation method using composite sensor nodes                              capability and can send a report Á . Here, Á is 1 if
       is derived from integral geometry and geometric prob-                              it detects the target object and 0 if otherwise. That is,
       ability, which are useful tools for analyzing a geometric                          Á    ½´ ´Ü Ý µ Ì              µ, where ½´ µ is an indicator
       or spatial structure. Ubiquitous access networks, such                             function that becomes 1 if a statement is true and 0 if
       as wireless sensor networks, require analysis of the                               otherwise. At Ø Ø , the -th sensor sends a report, Á ´Ø µ,
       geometric or spatial structure of an object, and there is                          describing whether or not it detects the target object. The
       literature (including one of our papers [11]) describing                           sensor network operator receives the report from each
       the use of integral geometry and geometric probability                             sensor through the network.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING

                                                                                                                                                                                    3



          Sensors are classified into multiple types according to                                             3     C OMPOSITE           SENSOR NODE
       their sensing areas. That is, different types of sensors                                              Estimating the unknown parameters of multiple target
       have different sensing area sizes or shapes. (A different                                             objects is difficult. Even when the number of target
       sensing area can often be implemented by changing the                                                 objects is known, a straight-forward extension of the
       sensitivity parameter of a sensor.) Type- sensors, of                                                 estimation method for a single target object to one for
       which sensing area is denoted by        , are deployed with                                           multiple target objects is not applicable if the parameters
       mean density . Assume that the first to Ò½ -th sensors                                                 of each target object are different. To estimate the pa-
       are type-1 sensors, the ´Ò½ · ½µ-th to Ò¾ -th sensors are                                             rameters of each target object, we introduce a composite
       type-2 sensors, ... In general, Ò  ½ · ½ ¡ ¡ ¡ Ò -th sensors
       are type- sensors. Á
                                       ØÑ
                                       Ø Ø½
                                              Ò    È È
                                                 Ò  ½ ·½ Á ´Øµ Ñ de-
                                                                                                             sensor node, which consists of multiple sensors arranged
                                                                                                             in a predetermined layout, such as a line. Instead of
       notes the time average of the number of type- sensors                                                 deploying individual sensors, the sensor network op-
       detecting the target object at each measurement epoch.                                                erator deploys composite sensor nodes. Although the
       We assume that the sensor network operator knows the                                                  locations of the composite sensor nodes are unknown,
       sensor type of each sensor sending a report.                                                          local and relative information among multiple sensors
          Consider a target object Ì in 2-dimensional space ʾ .                                             in a predetermined layout of a composite sensor node
       Its size is Ì , and its perimeter length is Ì . (In the                                               becomes available because the relative locations of these
       remainder of this paper, Ü denotes the size of Ü and                                                  sensors are known.
        Ü denotes the perimeter length of Ü.)
          Define a detectable area         such that sensors will                                             3.1 Straight-forward extension to multiple target ob-
       detect the target object if and only if they are located                                              jects
       in a detectable area       with a certain rotation angle.
       That is, ½´ ´Ü Ý µ        Ì        µ      ½ if and only if
                                                                                                             Before introducing composite sensor nodes, we explain
       ´Ü Ý µ ¾
                                                                                                             an extension of Eq. (1) to an equation applicable to
                     . Equivalently,           ´Ü Ý µ ´Ü Ý µ
       Ì           ÊÊÊ
                 . The detectable area size           is defined as
                                                                                                             multiple target objects.
                                                                                                             Proposition 1: Assume that there are Ò Ì convex and
                     ÜÝ ¾
                              Ü Ý    ¾ . We should note that
                              and Ì . When
                                                                                                             bounded target objects and that the sensing area     is
       depends on both                           is disk-shaped,
                                                                                                             also bounded and convex. Let    be the detectable areas
                                                                                                             of the -th target object and £
       does not depend on . When            does not depend on ,
                         ´Ü Ý µ ´Ü Ý µ Ì
                                                                                                                                                   . If
       we define                                    for simplicity.
                                                                                                             for any    ,
                                                                                                                                    Ì
                                                                                                                                    Ò                      Ò   Ì
       2.2       Previous results for basic model
                                                                                                                     £      ½
                                                                                                                                ´           Ì µ¡       ·           Ì   · ÒÌ       (5)
                                                                                                                           ¾
                                                                                                                                        ½                      ½
       In this section, we summarize our previous results,
       which are used in the current work. For the convex target
                                                                                                               This is because
                                                                                                                                                   È
                                                                                                             where Ì denotes the -th target object ´½
                                                                                                                                                  and because
                                                                                                                                                              ÒÌ µ. £
       object, we have obtained the following equation, and a
       shape and size estimation method was developed based                                                  satisfies Eq. (1).
                                                                                                               Therefore, we can estimate
                                                                                                                                              ÒÌ
                                                                                                                                                ½ Ì and
                                                                                                                                                       È      ÒÌ
                                                                                                                                                                 ½ Ì
                                                                                                                                                                              È
       on this equation [11]: Assume that both the target object
       Ì and the sensing area are bounded and convex. Then,                                                  instead of Ì and Ì in Eqs. (3) and (4), respectively.
                                                                                                             Consequently, if we can assume that all the target objects
                                      ½
                                      ¾
                                            Ì ¡            ·   Ì       ·                               (1)
                                                                                                             have the same shape and size, we can estimate the size
                                                                                                             and shape (size and perimeter length) of each target
                                                                                                             object. However, if not, we cannot estimate the size and
          Based on the fact that Á                 , we developed                                            shape (size and perimeter length) of each target object
       the following estimation method. (1) At Ø , receive the re-
                                          È È
                                                                                                             no matter how many types of sensors we introduce.
       port Á ´Ø µ from each sensor whose location is unknown.
                                Ñ     Ò
       (2) Calculate Á             ½    Ò  ½ ·½ Á ´Ø µ Ñ for each
       sensor type (     ½ ¾). (3) For two unknown parameters
                                                                                                             3.2   Extended model (static submodel)
        Ì and Ì , use Eq. (1) with the estimator Á of                                                        For analyzing an estimation method for multiple hetero-
       and solve the following equations for         ½ ¾.                                                    geneous target objects, we introduce a new model.
                                                                                                               A sensor network operator deploys composite sensor
                        ´
                            ½
                            ¾
                                  Ì ¡          ·       Ì       ·               µ       Á               (2)   nodes in a convex area ¨ in 2-dimensional space ʾ , but
                                                                                                             the operator does not know their locations. A composite
       where       Ì Ì           are estimators of                 Ì Ì             . That is,
                                                                                                             sensor node consists of multiple sensors arranged in a
                                                                                                             predetermined layout. We now consider a two-sensor

             Ì           ¾ ´Á½        ½       Á¾       ¾           ½       ·        ¾ µ
                                                                                                             composite sensor node, i.e., a composite sensor node
                                                   ½           ¾
                                                                                                       (3)   with two sensors. These two sensors are the same as
                                                                                                             those described in the basic model. That is, they are
           Ì                    ¾ ´     ½      Á½          ½µ ·        ½ ´         ¾         Á¾   ¾µ         simple binary sensors. For simplicity, in the remainder
                                                           ¾           ½
                                                                                                       (4)
                                                                                                             of this paper, we assume that the sensing area of each
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING

                                                                                                                                                                        4



       individual sensor in a composite sensor node is disk-                              4  A NALYSIS FOR EXTENDED MODEL                                   ( STATIC
       shaped.                                                                            SUBMODEL )
                                                                                          4.1 General results
          There are multiple types of composite sensor nodes.
       Type- composite sensor nodes are randomly deployed                                 In this subsection, we provide general measure proposi-
       with mean density          ¼. Each of the two sensors in a                         tions that only one (both) of the sensors in a composite
       type- composite sensor node has a disk-shaped sensing                              sensor node detects a target object. Let ѽ ´Ì µ be the
       area with radius Ö and is located at each of the two end                           measure of a set of composite sensor nodes in which one
       points of a line segment of length Ð . Assume that the first                        of the two sensors in each composite sensor node detects
       to Ò½ -th composite sensor nodes are type-1 composite                              a target object Ì and the other does not. (According to
       sensor nodes and the ´Ò½ · ½µ-th to Ò¾ -th composite
                                                                                          ѽ ´Ì µ
                                                                                                         Ê
                                                                                          Appendix A, ѽ ´Ì µ can be written by an integral form:
                                                                                                                              Ø.) Roughly, ѽ ´Ì µ is a
       sensor nodes are type-2 composite sensor nodes, ... In                                         ´Ü½ ܾ µ¾ ×       Ô
       general, Ò  ½ · ½ ¡ ¡ ¡ Ò -th composite sensor nodes are                           non-normalized probability that one of the two sensors
       type- composite sensor nodes for           ½ ¡ ¡ ¡ Â , where                       in a composite sensor node detects a target object Ì and
       Â is the number of composite sensor node types. (For                               the other does not, where the normalizing constant is Ñ
        ½     ¾, Ð ½  Ð ¾ or Ö ½  Ö ¾ .)                                                  (the measure of a set of composite sensor nodes placed

          Let Á be the report of the -th sensor of the -th                                A. Similarly, let Ѿ ´Ì µ
                                                                                                                                Ê
                                                                                          in (included in or intersects with) ¨), given in Appendix
                                                                                                                        ´Ü½ ܾ µ¾        Ô    Ø be the
       composite sensor node for           ½ ¾, and Á is 1 if it                          measure of a set of composite sensor nodes in which
       È
       detects the target object and 0 if otherwise. Let Á¾
          Ò
            Ò  ½ ·½ ½´Á½     Á¾      ½µ be the number of type-
                                                                                          both sensors of each composite sensor node detect a
                                                                                          target object Ì . Ѿ ´Ì µ is a non-normalized probability
       composite sensor nodes in each of which two sensors                                that both of the two sensors in a composite sensor node
                                              Ò       È
       detect at least one target object at a single measurement                          detect a target object Ì . (When we need to indicate
       epoch. Similarly, define Á½               Ò  ½ ·½ ½´ Á½ ´½                          the parameters Ð Ö of the composite sensor node to
       Á¾ µ · ´½   Á½ µÁ¾         ½µ, which denotes the number                            evaluate the measures, we use the notations ѽ ´Ì Ð Öµ
       of type- composite sensor nodes in each of which one                               and Ѿ ´Ì Ð Öµ. When we need to indicate the parameters
       of two sensors detects at least one target object at a single                      of the -th target object in Ѿ for the type- composite
       measurement epoch.                                                                 sensor node, we use the notation Ѿ ´¢ Ð Ö µ instead
                                                                                          of Ѿ ´Ì Ð Ö µ, where ¢ is defined later.)
          There are ÒÌ target objects in 2-dimensional space                              Proposition 2: Let ѽ ´ µ be the measure of a set of type-
       ʾ , where ÒÌ    is known. Let Ì be the -th target object                            composite sensor nodes in which only one of the two
       (½        ÒÌ ). (We propose the stochastic geometric filter,                        sensors in each composite sensor node detects any target
       which can estimate the number of target objects [21].) In                          object, and let Ѿ ´ µ be the measure of a set of type-
       this static submodel, target objects do not move.                                  composite sensor nodes in which both sensors of each
                                                                                          composite sensor node detect any target object. If there
          For type-      composite sensor nodes, define a                                  is no overlap between composite-detectable areas ½
       composite-detectable area         such that the sensors of                                   for any ½ ¾ (½       ½ ¾      ÒÌ ),
                                                                                             ¾
       the composite sensor nodes will detect the -th target                                                                   ÒÌ
       object if and only if they are located in a composite-
                                                                                                             ѽ ´   µ                   ѽ ´Ì Ð   Ö   µ              (6)
       detectable area            Ê . That is, when the locations                                                                   ½
       of the two sensors of a composite sensor node at the                                                                    ÒÌ
       two end points of a line segment of length Ð are Ü and                                                Ѿ ´   µ                   Ѿ ´Ì Ð   Ö   µ              (7)
       their disk-shaped sensing areas are ´Ü µ (                ½ ¾),                                                              ½
       ½´´ ´Ü½ µ Ì µ ´ ´Ü¾ µ Ì µ             µ     ½ if and only if                       £
       ´Ü½ ܾ µ ¾       , where ܽ   ܾ ¾         о . Similarly, we                        This is because, if there is no overlap of detectable
                                          ×
       define a single detectable area         (a double detectable                        areas, then the event that one of the two sensors (both
       area      ) such that only one (both) of the two sensors                           of the two sensors) in a composite sensor node detects
       of type- composite sensor nodes will detect the -th                                Ì or Ì is the sum of the two events. One is the event
       target object if and only if they are located in a single                          that one of the two sensors (both of the two sensors) in
       detectable area ×             Ê (in a double detectable                            the composite sensor node detects Ì , and the other is
       area           Ê ). That is, when the locations of the                             the event that one of the two sensors (both of the two
       two sensors of a composite sensor node at the two                                  sensors) in the composite sensor node detects Ì .
       end points of a line segment of length Ð are Ü and                                   We can derive other measures on this basis. For ex-
       their disk-shaped sensing areas are ´Ü µ (                ½ ¾),                    ample, the measure for at least one sensor in a type-
       ½´ ´Ü½ µ Ì          µ ¡ ½´ ´Ü¾ µ Ì       µ      ½ if and only                        composite sensor node detecting a target object is
       if ´Ü½ ܾ µ ¾       , where ܽ   ܾ ¾        о . In addition,                     ѽ ´ µ · Ѿ ´ µ.
         ×
                          . (We may remove        in       , × ,                          Remark 1: If there are overlaps of detectable areas of
       to simplify the notation, when        are not specified.)                           individual target objects (i.e., if ½      ¾
                                                                                                                                            ), the
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING

                                                                                                                                                                        5



       estimation error may increase (see Section 7. Numerical                            4.2.1     Definitions and notations
       examples). To avoid overlap, small sensing areas are                               Here, we provide a list of definitions and notations used
       preferable. In a large sensing area, it may not be able                            in this subsection.
       to detect a small gap between two target objects. This is
                                                                                            ¯ ¢ : a vector of parameters describing the -th target
       similar to the phenomenon of a large-sensing-area sensor
                                                                                               object.
       not detecting a small hole or a small concave part of a
                                                                                            ¯ ÒÔ ´ µ: the number of parameters in ¢ . (For simplic-
       target object, causing a large estimation error [11], [13].
                                                                                               ity, ÒÔ ´ µ is constant in the following if not explicitly
          For no overlaps between composite-detectable areas,
                                                                                               indicated.)
       two conditions are required. The first is that a single
       sensor in a composite sensor node should not simulta-                                ¯ ¢´ µ          ´¢ ¡ ¡ ¡ ¢ µ.
       neously detect multiple target objects. The second is that                           ¯ ¢       ¢´½ ÒÌ µ.
       the two sensors in a composite sensor node should not                                ¯   Û       ´Ð Ö µ.
       simultaneously detect multiple target objects. The first                              ¯   Ï Û ÈÛ  ´ ½ ¡ ¡ ¡ Â µ.
       condition is identical to that in which the detectable
       areas of individual target objects for a simple sensor
                                                                                                Ú Û
                                                                                            ¯ ´¢ µ                  ÒÌ
                                                                                                                       ½ Ѿ ´¢     µ.   Û
       do not overlap, and is also required in Proposition 1,                                   Ú Ï
                                                                                            ¯ ´¢ µ  Ú Û                              Ú Û
                                                                                                               ´ ´¢ ½ µ ¡ ¡ ¡ ´¢ Â µµ.
       which does not use a composite sensor node. However,
       the second condition is not needed when we do not use
                                                                                          4.3     Definition and proposition of observability
       the composite sensor node. This can be a new cause of
       errors, although composite sensor nodes can obtain new                             Definition 1: A value vector ¢ of parameter vector ¢ is
       information. To reduce this error, a shorter Ð is better.                          observable if there exists a set of composite sensor node
       Both conditions can be easily satisfied when the target                             parameter values           Ï  ´Ð½ Ö½ µ ¡ ¡ ¡ ´Ð Ö µ satisfying
       object’s density is low. £                                                               Ú Ï
                                                                                          that ´¢ µ              Ú
                                                                                                                ´¢¼      Ï
                                                                                                                       µ for any ¢¼       ¢ ¢¼ ¾ ËÔ for a
          The following proposition means that the expected                               given feasible parameter space Ë Ô . Here,                Ï
                                                                                                                                                is called an
       number of type- composite sensor nodes in which only                               observing parameter set. £
       one (both) of the two sensors detects a target object is                              Under an ideal situation (that is, there are no
       proportional to ѽ ´ µ (Ѿ ´ µ). This is natural because                           approximation or measurement errors), ´¢ µ                    Ú Ï
       of the definition of ѽ ´ µ (Ѿ ´ µ). See Appendix A for                              ´Á¾ ½ ¡ ¡ ¡ Á¾  µ. Therefore, roughly speaking, Definition
       mathematical details. In addition, the following propo-                            1 implies that if obtained sensor reports can uniquely
       sition is valid for any shaped target object.                                      determine ¢ under an ideal situation when we use a
       Proposition 3: Let Æ ´½ µ be the number of type-                                   certain set of composite sensor node parameter values
       composite sensor nodes in which one of the two sensors                             Ï  , ¢ is said to be “observable.” The statement that
       detects a target object, and let Æ ´¾ µ be the number                              obtained sensor reports can uniquely determine ¢ means
       of type- composite sensor nodes in each of which two                               that any other value vector ¢¼ of parameter vector ¢ is
       sensors detect a target object. If the composite sensor                            not consistent with the obtained sensor reports.
       nodes are distributed in a sufficiently large area,                                    The definition of observability requires the uniqueness
                                                                                          of the parameter value vector that is consistent to sensor
                               Æ   ´½ µ              ѽ   ´ µ                     (8)     reports. However, it does not require the uniqueness of
                               Æ   ´¾ µ              Ѿ   ´ µ                     (9)     Ú Ï
                                                                                            ´¢ µ over the entire domain of ¢. In addition, an
                                                                                          observing parameter set can depend on ¢.
       £                                                                                     Thus, the following proposition is directly derived
          Precisely, Eqs. (8) and (9) are affected by the shape                           from the definition of observability.
       of ¨, the sensor-deployed area (Appendix A). However,                                                      Ú Ï
                                                                                          Proposition 4: If ´¢ µ is given and if ¢ is observable
       if the border effect (the number of composite sensor                               with an observing parameter set                   Ï
                                                                                                                                        , we can uniquely
       nodes intersecting the border of ¨) is small, they are                             and exactly estimate ¢. £
       independent of the shape of ¨. Practically, this is the                               Because ´¢ µ Ú Ï            ´Á¾ ½ ¡ ¡ ¡ Á¾  µ if there is no
       case.                                                                              approximation or measurement errors, we can uniquely
          Note that the sample of the random variable Æ ´½ µ                              and exactly estimate observable parameter values by
       is Á½ and that of Æ ´¾ µ is Á¾ , and that Æ ´½ µ                                   using an observing parameter set               Ï
                                                                                                                                        and sensor reports.
          Á½   and Æ ´¾ µ          Á¾  .                                                  Equivalently, if ¢ is observable, there exists an observing
                                                                                          parameter set, and we can uniquely and exactly estimate
       4.2 Observability                                                                  ¢ by using it and sensor reports if there is no approxi-
                                                                                          mation or measurement errors. On the other hand, even
       In the remainder of this section, target objects are as-                           for observable parameter values, if the parameters of
       sumed to be convex. Without loss of generality, we can                             composite sensor nodes are not appropriately chosen,
       assume that ½ ¡ ¡ ¡         ÒÌ and н   ¾Ö½    ¡ ¡ ¡ Р  ¾Ö .                    we may not be able to appropriately estimate them.
       Here,
                                           Ô
                is the diameter of the -th target object, that is                         Unfortunately, we cannot judge whether a given                   is    Ï
             Ñ Ü´Ü½ ݽ µ ´Ü¾ ݾ µ¾Ì    ´Ü½   ܾ µ¾ · ´Ý½   ݾ µ¾ .                        an observing parameter set without knowing values of
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING

                                                                                                                                                                        6



                                    Ï
       ¢ or cannot provide , which is an observing parameter                              4.4    Derivation of measures
       set for any values of ¢.                                                           In the following subsections, we derive the measures
          If ½       ¡¡¡       ÒÌ , we can get a simplified sufficient                      ѽ and Ѿ for a certain class of target objects (disk-
       condition for observability. This condition can be deter-                          shaped and rectangular target objects) as examples.
       mined by an individual target object.                                              (Consequently, through Eqs. (6), (7), (8), and (9), Ñ ½ and
                                                                                          Ѿ , Æ ´½ µ and Æ ´¾ µ can be obtained.) We first
                                        Ï
       Lemma 1: If there exists a set of composite sensor
       node parameter values                 ´Ð½ Ö½ µ ¡ ¡ ¡ ´ÐÒÔ ÒÌ ÖÒÔ ÒÌ µ              derive the measures ѽ and Ѿ for a disk-shaped target
       satisfying that  ½            д  ½µÒÔ ·½   ¾Ö´  ½µÒÔ ·½       ¡¡¡
                                                                                          object. Second, we derive measures for rectangular target
       д  ½µÒÔ ·ÒÔ   ¾Ö´  ½µÒÔ ·ÒÔ                for ½
                               Û                                Û
                                                                          ÒÌ
       and that ´Ñ¾ ´¢¼ ´  ½µÒÔ ·½ µ ¡ ¡ ¡ Ѿ ´¢¼ ´  ½µÒÔ ·ÒÔ µµ
                                                                                          objects. Finally, they are derived when there are disk-

                  Û                                Û
       ´Ñ¾ ´¢ ´  ½µÒÔ ·½ µ ¡ ¡ ¡ Ѿ ´¢ ´  ½µÒÔ ·ÒÔ µµ when ¢¼
                                                                                          shaped and rectangular target objects.

       ¢ ¢¼ ¾ ËÔ ´ µ for a given feasible parameter space ËÔ ´ µ                          4.4.1 Disk-shaped target objects
       for ½                ÒÌ , ¢ is observable with an observing
       parameter set       Ï  .£
       Proof: Assume that ¢¼ £ ¢ £ and ¢¼ ¢ for £
                                                                                          When a target object is disk-shaped, we can obtain
                                                                                          explicit formulas. We derive ѽ and Ѿ under the as-
                                                                                          sumption that there is a single target object whose radius
       ÒÌ .                                                                               is Ê, the radius of each sensing area in a composite
          Suppose that  ½             д  ½µÒÔ ·½   ¾Ö´  ½µÒÔ ·½      ¡¡¡
                                                                                          sensor node is Ö, and the distance between the two
       д  ½µÒÔ ·ÒÔ   ¾Ö´  ½µÒÔ ·ÒÔ               for ½                   ÒÌ .            sensors in the node is Ð.
       Consider the type- composite sensor nodes where
       ´ £   ½µÒÔ · ½                   ´ £   ½µÒÔ · ÒÔ . Note that, if                      From Appendix B,
               Û È
               £ , Ѿ ´¢        µ ¼Ò because               Ð   ¾Ö . There-                          ѽ ´Ê Ð Öµ
         È Ú Û
       fore, ´¢         Ú Û
                         µ             Ì Ñ ´¢ µ and ´¢¼
                                          £  ¾
                           ¼ µ for ´ £  ½µÒÔ ·½
                                                                        µ                            ¾ ¾´Ê · Öµ¾ × Ò ½ ´ ¾´Êзֵ µ
                                                                                                           Ô
              ÒÌ
                  £ Ѿ ´¢                                    ´ £  ½µÒÔ ·ÒÔ .                         ·Ð ´Ê · Öµ¾   о                                ¾´Ê · Öµ
            Ú Û Ú Û
       Hence, ´¢           µ   ´¢¼ µ ´Ñ¾ ´¢ £ µ   Ѿ ´¢¼ £ µµ.                                       ¾ ¾ ´Ê · Öµ¾
                                                                                                                                   for
                                                                                                                                   for               ¾´Ê · Öµ
                                                                                                                                                                     Ð
                                                                                                                                                                     Ð.


        Ú Û Ú Û Ú Û Ú Û
       According to the assumption of this lemma, for                                                                                                               (10)
       ¢¼ £ ¢ £ , ´ ´¢ ´ £  ½µÒÔ ·½ µ ¡ ¡ ¡ ´¢ ´ £  ½µÒÔ ·ÒÔ µµ                                     Ѿ ´Ê Ð Öµ
       ´ ´¢¼ ´ £  ½µÒÔ ·½ µ ¡ ¡ ¡ ´¢¼ ´ £  ½µÒÔ ·ÒÔ µµ.                                                  ´Ê Ð Ö µ       for ¾´Ê · Öµ
                Ú Ï Ú Ï
                                                                                                                                               Ð,
          Consequently, ´¢ µ                ´¢¼ µ if ¢¼ ¢. £                                        ¼                   otherwise,
                                                                                                                                                                    (11)
          In practice, we are likely to face the following situ-
       ation: The target object shape can be categorized into                             where ´Ê Ð Öµ
                                                                                           Ô                             ´Ê · Öµ¾ ´             ¾ × Ò ½ ´ ¾´Êзֵ µµ    

       several categories, such as disks and rectangles, and                              Ð ´ Ê · Ö µ¾ Ð ¾  .
       we may not know how many target objects belong to                                  Remark 2: When there are Ò Ì disk-shaped target objects
       each category. Note that ÒÔ ´ µ is likely to depend on the
                                                                                                                     Ï
                                                                                          and the radius Ê of the -th target object satisfies
       category to which the -th target object belongs. Let be                            ʽ      ¡¡¡   ÊÒÌ ,     that satisfies н   ¾Ö½     ¾Ê½
       the number of target objects in the -th category and Ò                             о   ¾Ö¾ ¾Ê¾ ¡ ¡ ¡ ÐÒÌ   ¾ÖÒÌ ¾ÊÒÌ is an observing
       the number of categories. ½ ¡ ¡ ¡ Ò are also unknown                               parameter set, due to Lemma 1. This is because ´Ê Ð Öµ
       parameters. Similar to Proposition 4, the following corol-                         is an increasing function of Ê. £
       lary shows that we can estimate ½ ¡ ¡ ¡ Ò as well as                                  In the remainder of this paper, if we need to explicitly
       other observable parameters ¢.
                             Ú Ï
                                                                                          indicate “disk-shaped target object” for these measures
       Corollary 1: If ´¢ µ is given and if ¢ and values of                               ѽ and Ѿ , we use the notations ѽ and Ѿ .
        ½ ¡ ¡ ¡ Ò are observable, we can uniquely and exactly
       estimate them. £                                                                   4.4.2 Rectangular target objects
          Proposition 4, Lemma 1, and the corollary mentioned
       above mean that if we can provide more than
                                                                          È
                                                                       ÒÔ ´ µ
                                                                                          This subsection analyzes rectangular target objects. Con-
                                                                                          sider a single rectangular target object with two sides
       types of composite sensor nodes with appropriate Ð Ö                               and a single type of composite sensor node whose sen-
       and a sufficiently large number of samples of sensed re-                            sors’ sensing-area radius is Ö and where the distance be-
       sults, we can estimate observable values of parameters of                          tween the sensors is Ð. The necessary and sufficient con-
       any convex target object by using two-sensor composite                             dition of the first (second) sensor in a composite sensor
       sensor nodes. To concretely obtain estimates, a calcu-                             node detecting the target object is that the location of the
       lation method for Ѿ ´¢ Ð Öµ is required. As examples,                             first (second) sensor is in . Here, is the detectable area
       we provide formulas to calculate Ѿ ´¢ Ð Öµ for a certain                          of this rectangular target object when a basic (i.e., non-
       class of target objects. Theoretically, a simulation is ap-                        composite) disk-shaped sensing area with a radius Ö is
       plicable by doing a simulation for various values of ¢                             used. That is,         ´Ü ݵ Ñ Ò  ¾ ܼ ¾   ¾ ݼ ¾ ´Ü 
       for each pair of ´Ð Ö µ to obtain Ѿ ´¢ Ð Ö µ. However,                            ܼ µ¾ · ´Ý   Ý ¼ µ¾    Ö¾ . To simplify the calculation, we
       practically, the applicability of the simulation is limited                        introduce           ´Ü ݵ   ¾   Ö Ü        ¾· Ö   ¾   Ö
       to special cases, for example, those in which the shapes                           Ý        ¾ · Ö instead of (Fig. 1). That is,               .
       of the target objects and the ranges of parameter values                           Then, the necessary and sufficient condition of the first
       are roughly known in advance.                                                      (second) sensor in a composite sensor node detecting the
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING

                                                                                                                                                                                               7



                                                                                                                       of ´Ð Ö µ, which satisfies Ð   ¾Ö      Ñ Ò´ µ is not
                     D

                                                                                                                                                                                             Ï
               D                                                                                                       included in an observing parameter set.
                                                                                                                         On the other hand, if ¾ · ¾
                                                                                                                                              Ô ½   ½    ¡¡¡
                                                                                                                                                                 ¾ · ¾ ,
                                                                                                                                                                 ÒÌ     ÒÌ
                                                                                                                       that satisfies Ñ Ü´ ´  ½ · ¾Ö¾ µ ¾·´
                                                                                                                                                               ½ · ¾Ö¾ µ¾ ·
                                                                                                                                                                     Ô · ·¾
                                                                         l
                                                                                                                       ¾Ö¾    · ¾Ö¾ µ         о  ½    о        ¾   ¾    Ö¾ Ö¾
                                                             t
                                                                 p
                                                                     H
                                                                                     r                                 Ö¾  ½ о    о  ½  ·   Æ is an observing parameter set where
                                                                                                                       Æ is a sufficiently small positive scalar. See Appendix D
                         b                                                                                             for details. £
                                                          a                                                              In the remainder of this paper, if we need to explicitly
                                                                                                                       indicate “rectangle” for these measures Ñ ½ and Ѿ , we
                                                                                                                       use the notations ѽ Ö and Ѿ Ö .
                                                     r                           G
                                                                                                                       4.4.3 Combinations of disk-shaped and rectangular tar-
                                                                                                                       get objects
                                                                                                                       Let Ò be the number of disk-shaped target objects and
       Fig. 1. Analysis of two-sensor composite sensor node for                                                        ÒÖ   ÒÌ   Ò be the number of rectangular target objects.
       rectangular target object                                                                                       Ò and ÒÖ are unknown parameters. As the measures are
                                                                                                                       additive if ½         ¾
                                                                                                                                                      for any ½ ¾ (½    ½ ¾
                                                                                                                       ÒÌ ½     ¾ ), we can easily obtain
       target object is approximately equivalent to the location                                                                                 Ò
       of the first (second) sensor being in . We use this                                                                       ѽ   ´µ                  ѽ   ´Ê     Ð Ö      µ
       approximation and derive measures. Because brute-force                                                                                        ½
                                                                                                                                                              ÒÖ
       but lengthy computations are needed, we show only the
       results here. The computation details are in Appendix C.                                                                                          ·         ѽ Ö   ´       Ð Ö   µ   (14)
           Define           · ¾Ö ,         · ¾Ö, « Ñ Ò´ µ, and                           Ò
                                                                                                                                                               ½

       ¬       Ñ Ü´ µ.                                                    Ѿ ´ µ            Ѿ ´Ê Ð Ö µ
                                                                                          ½
               ѽ ´    Ð Öµ                                                                    ÒÖ
                   д · µ ¾Ð     ¾              for Ð «,                                    · Ѿ Ö ´         Ð Ö µ     (15)
                   «¬ Ó×  ½ ´« е · Ь                                                           ½
                     ¬ о «¾ · «¾
                              Ô
                                                for « Ð ¬ ,      where Ê is the radius of the -th disk-shaped target ob-
                     ´Ô  ½ ´ Ð µ · Ó× ½ ´ Ð µµ
                      Ó×                                    (12) ject and       are the side lengths of the -th rectangular
                                                                                               Ô¾
                                                         Ô
                                  о   ¾                     о   ¾                                                    target object.
                   ·¾´ · · µ  ¾        ¾        о                           Ô ¾Ð
                                                                         for ¬                           ·        ¾,

                   ¾                                                     for             ·     ¾        Ð,             5     E XTENDED        MODEL ( DYNAMIC SUBMODEL )
                                                                                                                       5.1   Model description
                Ѿ   ´        Ð Ö  µ
                       · ¾´ · µÐ¾   Ð                                            for Ð             «,
                                                                                                                       The difference between the static and dynamic sub-
                   ¾«¬ ´ ¾ Ó× ½ ´« еµ ¾Ð¬
                                                                      
                                                                                                                       models is as follows: The target objects can move and
                   ·¾¬ о «¾ «¾
                              Ô
                                                                                                                       every composite sensor node sends a report at each
                                                                                 for «             Ð         ¬,
                   ¾ ´Ô ¾ Ó× ½ ´ Ð µ Ó× ½ ´ Ð µµ
                                                              
                                                                                                                       measurement epoch. There are no other differences.
                                                                                                                          More precisely, the dynamic submodel is as follows.
                   ·¾ о ¾ · ¾ о ¾
                                                         Ô
                                                                  
                                                                                                                       Each of the ÒÌ target objects may move along an
                    
                          ¾    
                                   ¾       о                                    for ¬   ԾР                          unknown route with unknown (maybe time-variant)
                                                                                         Ô ¾·           ¾,
                                                                                                                       speed. Every composite sensor node sends a report at
                   ¼                                                             for                ·        ¾
                                                                                                                       each measurement epoch. The -th sensor of the -th
                                                                                          Ð,                           composite sensor node sends the report Á ´Ø µ at time
                                                                                                        (13)
                                                                                                                       Ø , where         ½ ¾,      ½ ¡ ¡ ¡  . Redefine Á¾ as the
       Remark 3: When there are multiple rectangular target                                                            time average of the number of type- composite sensor
       objects with side lengths     (½         ÒÌ ) satisfy                                                           nodes in each of which two sensors detect at least
       Ð   ¾Ö       Ñ Ò´ µ, Eqs. (7), (13), and (9) show                                                                          È       È
                                                                                                                                                         Ƚ È
               Æ ´¾ µ
                                                 È
                                ´ ´ · ¾Ö µ´ · ¾Ö µ · о  
                                                                                                                       one target object at a single measurement epoch, that
                                                                                                                                     ØÑ     Ò
                                                                                                                                              Ò  ½ ·½ ´Á½ ´Øµ       Á¾ ´Øµ   ½µ Ñ.
                                                         ´ È                                 È´ ·                      is Á¾
       that
       ¾Ð ´ · · Ö µµ                  · ¾´ Ö                                             Ð µ
                                                                                                                                     Ø Ø½
                                                                                                                       Similarly, redefine Á½
                                                                                                                                                     ØÑ
                                                                                                                                                     Ø Ø½
                                                                                                                                                             Ò
                                                                                                                                                               Ò  ½ ·½            ½
                                                                                                                                                                       ´ Á½ ´Øµ´½  
         µ · ´ Ö¾ · о Ð Ö µÒÌ µ . Thus, if we can use
                                                                                                                       Á¾ ´Øµµ · ´½   Á½ ´ØµµÁ¾ ´Øµ        ½µ Ñ, which denotes the
          Æ ´¾ È withÈ
                µ    various Ð Ö simultaneously, we can es-                                                            time average of the number of type- composite sensor
       timate          ´ · µ. However, we cannot estimate                                                              nodes in each of which one of two sensors detects at
       each              . Therefore, it is often the case that a pair                                                 least one target object at a single measurement epoch.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING

                                                                                                                                                                                  8



       5.2     Analyzed results for dynamic submodel                                                             where ÈÈ     Ò
                                                                                                                                Ò
                                                                                                                                  Ñ ½ ´Ê Û½ µ
                                                                                                                                                    ÈÈ
                                                                                                                                       is an estimator of Ò ,
                                                                                                                                                       ÒÌ  Ò
                                                                                                                                                                       Ù¾
                                                                                                                                                                             Û½ µµ
                                                                                                                       ÈÈ                           ÈÈ
                                                                                                                 ´   ½´                         ·             ѽ Ö ´
                                                                                                                                                      ÒÌ  Ò
       It should be noted that the analyzed results originally                                                                  ½                          ½
                                                                                                                                       Ñ ½ ´Ê ÛÂ µ ·                        ÛÂ µµ
                                                                                                                                  Ò
                                                                                                                 ¡¡¡          ´                              ѽ Ö ´
                                                                                                                                                       ÒÌ  Ò
       derived for the static submodel are valid for the dy-                                                              Â          ½                   ½
                                                                                                                               ½ Ѿ ´Ê Û½ µ                                  Û½ µµ
                                                                                                                          Ò
       namic submodel. The reasons are as follows. (1) At each                                                       ½´                         ·             Ѿ Ö ´
                                                                                                                                                     ÒÌ  Ò
                                                                                                                                                           ½
                                                                                                                                       Ѿ ´Ê Û µ ·                         Û µµµ.
                                                                                                                                  Ò
       measurement epoch, the dynamic submodel is identical                                                      ¡¡¡      Â   ´
                                                                                                                                     ½                  ½    Ѿ Ö ´
       to the static submodel. (2) Only the quantity affected
       by multiple measurement epochs is Á , but it is no                                                        7        N UMERICAL      EXAMPLES
       included in derived formulas.          Æ´ µ         Á    is
                                                                                                                 This section provides numerical examples. The following
       valid both for the static and dynamic submodels. The
                                                                                                                 conditions were used as a basic pattern for the simula-
       fact Ú´¢ ϵ      ´Á¾ ½ ¡ ¡ ¡ Á¾  µ under the assumption of
                                                                                                                 tion. We used a monitored rectangular area, ¾¼ ¼¼¼ ¢ ½¼¼
       no approximation errors or measurement errors is also
                                                                                                                 square units, in which composite sensor nodes were
       valid.
                                                                                                                 deployed. Three target objects moved at a speed equal to
                                                                                                                 10 units of length per unit time along a straight line that
       6       E STIMATION                METHOD                                                                 was parallel to the bottom line of the monitored area.
       Based on the analysis in the previous section, we propose                                                 Two of the objects were disk-shaped with radiuses of 3
       an estimation method for multiple target objects that                                                     and 30, and the other one was rectangular with sides
       may have different parameters.                                                                            (3, 10). We used six composite sensor nodes of which
         Note that    Á          Æ ´ µ for        ½ ¾. Thus, Á¾                                                  parameters ´Ð Ö µ were (3, 1), (4, 1), (9, 2), (12, 3), (20,
       can be an estimator of Æ ´¾ µ .                                                                           2), and (22, 1) for ½        . We set      ¼    per square
                                                                                                                 unit length for all , and composite senor nodes were
                               Á½      ·     ½                      Æ ´½     µ                            (16)   placed in a homogeneous Poisson process. (As a result,
                               Á¾      ·     ¾                      Æ ´¾     µ                            (17)   the mean density of the sensors was 1 per square unit
                                                                                                                 length.) The mean distance between the target objects
       where            Á      Á    is an error of Á from its                                                    was 1,000. One simulation yields 2000 measurement
       expectation. By using Eqs. (6), (7), (8), and (9),                                                        epochs, and 10 simulation were run to obtain each result.
                                                          ÒÌ

                         Á½   ·    ½                            ѽ ´Ì Ð          Ö   µ                    (18)
                                                                                                                 7.1 Approximation errors and sensitivities to vari-
                                                           ½
                                                          ÒÌ                                                     ous conditions
                         Á¾   ·    ¾                            Ѿ ´Ì Ð          Ö   µ                    (19)   We first confirmed the agreement of the simulation re-
                                                           ½                                                     sults and the theoretically-derived results and evaluated
       The right-hand sides of these two equations are given                                                     approximation errors under various conditions and sen-
       by derived measures for each class of target object. For                                                  sitivities of Á½ (Á¾ ) to various conditions. (In 7.2.1 and
       example, if the target objects are disk-shaped (rectan-                                                   7.2.2, the impact of these conditions and errors on the
       gles), Eqs. (10) and (11) (Eqs. (12) and (13)) can be used.                                               estimation accuracy is shown.) We compared Á½ (Á¾ )



               ÈÈ                                    ÈÈ
       When there may be both disk-shaped and rectangular                                                        with Æ ´½ µ ( Æ ´¾ µ ), that is, the right-hand side
       objects, the right-hand sides of Eqs. (18) and (19) should                                                of Eq. (8) (Eq. (9)). For the disk-shaped target objects,
                  Ò                       ÒÖ
       be ´         ½ ѽ ´Ê Ð Ö µ ·          ½ ѽ Ö ´    Ð Ö µµ ,                                                Eqs. (10) and (11) were used, and for the rectangular
             Ò                       ÒÖ
         ´
                ½ Ѿ ´Ê Ð Ö µ ·         ½ Ѿ Ö ´      Ð Ö µµ .                                                   target object, Eqs. (12) and (13) were used.
         In general,
                                                                                                                 7.1.1 Basic pattern
                                          Á·                    Ù                                         (20)
                                                                                                                 For the basic pattern, Figure 2 shows the relative errors
                     Á                 ´Á½   ½ Á½ ¾             Á½         Á¾ ½ Á¾ ¾                 Á¾  µ,    of the theoretical values (that is, the relative error =

               È                                          È
       where                                              ¡¡¡                                  ¡¡¡
                                                                                                                 theoretical value/simulation result -1). Æ ´½ µ shows
                 ´   ½½       ½¾    ¡¡¡      ½   Â    ¾½        ¾¾    ¡¡¡        ¾   Â µ  , and           Ù
                                                                                                                 a positive bias because we approximated           by    for
       ´   ½
                ÒÌ
                     ½ ѽ ´¢        Û½ µ     ¡¡¡      Â
                                                                ÒÌ
                                                                     ½ ѽ ´¢         Û        µ   Ú µ.
                                                                                                                 the rectangular target object (see Fig. 1). Æ ´¾ µ also
           A set of that minimizes the square error Ì ´Á  
                          ¢
                                                                                                                 can have a positive bias, but it was within a range
       Ù µ´Á   Ù µÌ can be an estimator ¢ of ¢, where Ì is
                                                                                                                 of simulation error (see Figure 4 for the variance of
       a transpose operator.
                                                                                                                 the simulation results). In total, the relative errors were
                          ¢         Ö Ñ Ò¢ ´         Á Ù         µ´   Á Ù            µ
                                                                                      Ì
                                                                                                          (21)   small, and we concluded that the theoretical results are
                                                                                                                 valid.
       When the target object shape can be categorized into sev-
       eral categories, such as disks and rectangles, the number                                                 7.1.2 Independence to speed, monitored area, and
       of target objects in each category is also a parameter to                                                 moving directions
       be estimated. For example, when there may be both disk-
       shaped and rectangular objects,                                                                           Fig. 3 provides Á¾ when one condition such as the target
                                                                                                                 object speed is modified among conditions used in the
               ´¢    Ò    µ       Ö Ñ Ò¢ Ò ´         Á   Ù¾           µ´   Á   Ù¾          µ
                                                                                            Ì
                                                                                                          (22)   basic pattern.
Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes
Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes
Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes
Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes
Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes
Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes

More Related Content

What's hot

NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
ijwmn
 
Energy efficient protocol in wsn WITH ACO
Energy efficient protocol in wsn WITH ACOEnergy efficient protocol in wsn WITH ACO
Energy efficient protocol in wsn WITH ACO
Neelam Choudhary
 
Localization with mobile anchor points in wireless sensor networks
Localization with mobile anchor points in wireless sensor networksLocalization with mobile anchor points in wireless sensor networks
Localization with mobile anchor points in wireless sensor networks
Habibur Rahman
 
Significant Storage on Sensor Storage Space, Energy Consumption and Better Se...
Significant Storage on Sensor Storage Space, Energy Consumption and Better Se...Significant Storage on Sensor Storage Space, Energy Consumption and Better Se...
Significant Storage on Sensor Storage Space, Energy Consumption and Better Se...
ijasuc
 
29. continuous neighbor discovery in asynchronous sensor networks
29. continuous neighbor discovery in asynchronous sensor networks29. continuous neighbor discovery in asynchronous sensor networks
29. continuous neighbor discovery in asynchronous sensor networks
akshuu16
 

What's hot (16)

Ka2417181721
Ka2417181721Ka2417181721
Ka2417181721
 
International Journal of Engineering Research and Development (IJERD)
International Journal of Engineering Research and Development (IJERD)International Journal of Engineering Research and Development (IJERD)
International Journal of Engineering Research and Development (IJERD)
 
Multidimensional scaling algorithm and its current applications in wireless l...
Multidimensional scaling algorithm and its current applications in wireless l...Multidimensional scaling algorithm and its current applications in wireless l...
Multidimensional scaling algorithm and its current applications in wireless l...
 
Concealed Data Aggregation with Dynamic Intrusion Detection System to Remove ...
Concealed Data Aggregation with Dynamic Intrusion Detection System to Remove ...Concealed Data Aggregation with Dynamic Intrusion Detection System to Remove ...
Concealed Data Aggregation with Dynamic Intrusion Detection System to Remove ...
 
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
 
Energy efficient protocol in wsn WITH ACO
Energy efficient protocol in wsn WITH ACOEnergy efficient protocol in wsn WITH ACO
Energy efficient protocol in wsn WITH ACO
 
Localization with mobile anchor points in wireless sensor networks
Localization with mobile anchor points in wireless sensor networksLocalization with mobile anchor points in wireless sensor networks
Localization with mobile anchor points in wireless sensor networks
 
Sensor Deployment Algorithm for Hole Detection and Healing By Using Local Hea...
Sensor Deployment Algorithm for Hole Detection and Healing By Using Local Hea...Sensor Deployment Algorithm for Hole Detection and Healing By Using Local Hea...
Sensor Deployment Algorithm for Hole Detection and Healing By Using Local Hea...
 
F33022028
F33022028F33022028
F33022028
 
F010524057
F010524057F010524057
F010524057
 
Significant Storage on Sensor Storage Space, Energy Consumption and Better Se...
Significant Storage on Sensor Storage Space, Energy Consumption and Better Se...Significant Storage on Sensor Storage Space, Energy Consumption and Better Se...
Significant Storage on Sensor Storage Space, Energy Consumption and Better Se...
 
International Journal of Engineering Research and Development
International Journal of Engineering Research and DevelopmentInternational Journal of Engineering Research and Development
International Journal of Engineering Research and Development
 
29. continuous neighbor discovery in asynchronous sensor networks
29. continuous neighbor discovery in asynchronous sensor networks29. continuous neighbor discovery in asynchronous sensor networks
29. continuous neighbor discovery in asynchronous sensor networks
 
localization in wsn
localization in wsnlocalization in wsn
localization in wsn
 
Parallel and-distributed-system-ieee-2014-projects
Parallel and-distributed-system-ieee-2014-projectsParallel and-distributed-system-ieee-2014-projects
Parallel and-distributed-system-ieee-2014-projects
 
Parallel and Distributed System IEEE 2014 Projects
Parallel and Distributed System IEEE 2014 ProjectsParallel and Distributed System IEEE 2014 Projects
Parallel and Distributed System IEEE 2014 Projects
 

Viewers also liked

A novel estimation based backoff algorithm in the ieee
A novel estimation based backoff algorithm in the ieeeA novel estimation based backoff algorithm in the ieee
A novel estimation based backoff algorithm in the ieee
ambitlick
 
Multi hop wireless-networks
Multi hop wireless-networksMulti hop wireless-networks
Multi hop wireless-networks
ambitlick
 
An adaptive power controlled mac protocol for
An adaptive power controlled mac protocol forAn adaptive power controlled mac protocol for
An adaptive power controlled mac protocol for
ambitlick
 
Effective and efficient shape based pattern
Effective and efficient shape based patternEffective and efficient shape based pattern
Effective and efficient shape based pattern
ambitlick
 
Dynamic framed slotted aloha algorithms using fast tag estimation
Dynamic framed slotted aloha algorithms using fast tag estimationDynamic framed slotted aloha algorithms using fast tag estimation
Dynamic framed slotted aloha algorithms using fast tag estimation
ambitlick
 
Label based dv-hop localization against wormhole attacks in wireless sensor n...
Label based dv-hop localization against wormhole attacks in wireless sensor n...Label based dv-hop localization against wormhole attacks in wireless sensor n...
Label based dv-hop localization against wormhole attacks in wireless sensor n...
ambitlick
 

Viewers also liked (8)

A novel estimation based backoff algorithm in the ieee
A novel estimation based backoff algorithm in the ieeeA novel estimation based backoff algorithm in the ieee
A novel estimation based backoff algorithm in the ieee
 
Multi hop wireless-networks
Multi hop wireless-networksMulti hop wireless-networks
Multi hop wireless-networks
 
An adaptive power controlled mac protocol for
An adaptive power controlled mac protocol forAn adaptive power controlled mac protocol for
An adaptive power controlled mac protocol for
 
Independent tasks scheduling based on genetic
Independent tasks scheduling based on geneticIndependent tasks scheduling based on genetic
Independent tasks scheduling based on genetic
 
Effective and efficient shape based pattern
Effective and efficient shape based patternEffective and efficient shape based pattern
Effective and efficient shape based pattern
 
Backbone nodes based stable routing for mobile ad hoc networks
Backbone nodes based stable routing for mobile ad hoc networksBackbone nodes based stable routing for mobile ad hoc networks
Backbone nodes based stable routing for mobile ad hoc networks
 
Dynamic framed slotted aloha algorithms using fast tag estimation
Dynamic framed slotted aloha algorithms using fast tag estimationDynamic framed slotted aloha algorithms using fast tag estimation
Dynamic framed slotted aloha algorithms using fast tag estimation
 
Label based dv-hop localization against wormhole attacks in wireless sensor n...
Label based dv-hop localization against wormhole attacks in wireless sensor n...Label based dv-hop localization against wormhole attacks in wireless sensor n...
Label based dv-hop localization against wormhole attacks in wireless sensor n...
 

Similar to Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes

1 Object tracking using sensor network Orla Sahi
1       Object tracking using sensor network Orla Sahi1       Object tracking using sensor network Orla Sahi
1 Object tracking using sensor network Orla Sahi
SilvaGraf83
 
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
ijwmn
 
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
ijwmn
 
Ijarcet vol-2-issue-2-756-760
Ijarcet vol-2-issue-2-756-760Ijarcet vol-2-issue-2-756-760
Ijarcet vol-2-issue-2-756-760
Editor IJARCET
 
A self localization scheme for mobile wireless sensor networks
A self localization scheme for mobile wireless sensor networksA self localization scheme for mobile wireless sensor networks
A self localization scheme for mobile wireless sensor networks
ambitlick
 

Similar to Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes (20)

A2 l
A2 lA2 l
A2 l
 
Ii2414621475
Ii2414621475Ii2414621475
Ii2414621475
 
Bi24385389
Bi24385389Bi24385389
Bi24385389
 
F0361026033
F0361026033F0361026033
F0361026033
 
1 Object tracking using sensor network Orla Sahi
1       Object tracking using sensor network Orla Sahi1       Object tracking using sensor network Orla Sahi
1 Object tracking using sensor network Orla Sahi
 
wireless sensor network
wireless sensor networkwireless sensor network
wireless sensor network
 
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
 
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
NETWORK PERFORMANCE ENHANCEMENT WITH OPTIMIZATION SENSOR PLACEMENT IN WIRELES...
 
AN IMPROVED DECENTRALIZED APPROACH FOR TRACKING MULTIPLE MOBILE TARGETS THROU...
AN IMPROVED DECENTRALIZED APPROACH FOR TRACKING MULTIPLE MOBILE TARGETS THROU...AN IMPROVED DECENTRALIZED APPROACH FOR TRACKING MULTIPLE MOBILE TARGETS THROU...
AN IMPROVED DECENTRALIZED APPROACH FOR TRACKING MULTIPLE MOBILE TARGETS THROU...
 
F33022028
F33022028F33022028
F33022028
 
Node Deployment in Homogeneous and Heterogeneous Wireless Sensor Network
Node Deployment in Homogeneous and Heterogeneous Wireless Sensor NetworkNode Deployment in Homogeneous and Heterogeneous Wireless Sensor Network
Node Deployment in Homogeneous and Heterogeneous Wireless Sensor Network
 
Design Issues and Applications of Wireless Sensor Network
Design Issues and Applications of Wireless Sensor NetworkDesign Issues and Applications of Wireless Sensor Network
Design Issues and Applications of Wireless Sensor Network
 
Accurate and Energy-Efficient Range-Free Localization for Mobile Sensor Networks
Accurate and Energy-Efficient Range-Free Localization for Mobile Sensor NetworksAccurate and Energy-Efficient Range-Free Localization for Mobile Sensor Networks
Accurate and Energy-Efficient Range-Free Localization for Mobile Sensor Networks
 
Ijarcet vol-2-issue-2-756-760
Ijarcet vol-2-issue-2-756-760Ijarcet vol-2-issue-2-756-760
Ijarcet vol-2-issue-2-756-760
 
DATA TRANSMISSION IN WIRELESS SENSOR NETWORKS FOR EFFECTIVE AND SECURE COMMUN...
DATA TRANSMISSION IN WIRELESS SENSOR NETWORKS FOR EFFECTIVE AND SECURE COMMUN...DATA TRANSMISSION IN WIRELESS SENSOR NETWORKS FOR EFFECTIVE AND SECURE COMMUN...
DATA TRANSMISSION IN WIRELESS SENSOR NETWORKS FOR EFFECTIVE AND SECURE COMMUN...
 
A Novel Three-Dimensional Adaptive Localization (T-Dial) Algorithm for Wirele...
A Novel Three-Dimensional Adaptive Localization (T-Dial) Algorithm for Wirele...A Novel Three-Dimensional Adaptive Localization (T-Dial) Algorithm for Wirele...
A Novel Three-Dimensional Adaptive Localization (T-Dial) Algorithm for Wirele...
 
J017345864
J017345864J017345864
J017345864
 
A self localization scheme for mobile wireless sensor networks
A self localization scheme for mobile wireless sensor networksA self localization scheme for mobile wireless sensor networks
A self localization scheme for mobile wireless sensor networks
 
S04404116120
S04404116120S04404116120
S04404116120
 
E010412433
E010412433E010412433
E010412433
 

More from ambitlick

Ambitlick ns2 2013
Ambitlick ns2 2013Ambitlick ns2 2013
Ambitlick ns2 2013
ambitlick
 
Low cost Java IEEE Projects 2013
Low cost Java IEEE Projects 2013Low cost Java IEEE Projects 2013
Low cost Java IEEE Projects 2013
ambitlick
 
Handling selfishness in replica allocation
Handling selfishness in replica allocationHandling selfishness in replica allocation
Handling selfishness in replica allocation
ambitlick
 
Mutual distance bounding protocols
Mutual distance bounding protocolsMutual distance bounding protocols
Mutual distance bounding protocols
ambitlick
 
Moderated group authoring system for campus wide workgroups
Moderated group authoring system for campus wide workgroupsModerated group authoring system for campus wide workgroups
Moderated group authoring system for campus wide workgroups
ambitlick
 
Efficient spread spectrum communication without pre shared secrets
Efficient spread spectrum communication without pre shared secretsEfficient spread spectrum communication without pre shared secrets
Efficient spread spectrum communication without pre shared secrets
ambitlick
 
Integrated institutional portal
Integrated institutional portalIntegrated institutional portal
Integrated institutional portal
ambitlick
 
Moderated group authoring system for campus wide workgroups
Moderated group authoring system for campus wide workgroupsModerated group authoring system for campus wide workgroups
Moderated group authoring system for campus wide workgroups
ambitlick
 

More from ambitlick (20)

DCIM: Distributed Cache Invalidation Method for Maintaining Cache Consistency...
DCIM: Distributed Cache Invalidation Method for Maintaining Cache Consistency...DCIM: Distributed Cache Invalidation Method for Maintaining Cache Consistency...
DCIM: Distributed Cache Invalidation Method for Maintaining Cache Consistency...
 
Low cost Java 2013 IEEE projects
Low cost Java 2013 IEEE projectsLow cost Java 2013 IEEE projects
Low cost Java 2013 IEEE projects
 
Ambitlick ns2 2013
Ambitlick ns2 2013Ambitlick ns2 2013
Ambitlick ns2 2013
 
Low cost Java IEEE Projects 2013
Low cost Java IEEE Projects 2013Low cost Java IEEE Projects 2013
Low cost Java IEEE Projects 2013
 
Handling selfishness in replica allocation
Handling selfishness in replica allocationHandling selfishness in replica allocation
Handling selfishness in replica allocation
 
Mutual distance bounding protocols
Mutual distance bounding protocolsMutual distance bounding protocols
Mutual distance bounding protocols
 
Moderated group authoring system for campus wide workgroups
Moderated group authoring system for campus wide workgroupsModerated group authoring system for campus wide workgroups
Moderated group authoring system for campus wide workgroups
 
Efficient spread spectrum communication without pre shared secrets
Efficient spread spectrum communication without pre shared secretsEfficient spread spectrum communication without pre shared secrets
Efficient spread spectrum communication without pre shared secrets
 
IEEE -2012-13 Projects IN NS2
IEEE -2012-13 Projects IN NS2  IEEE -2012-13 Projects IN NS2
IEEE -2012-13 Projects IN NS2
 
Adaptive weight factor estimation from user review 1
Adaptive weight factor estimation from user   review 1Adaptive weight factor estimation from user   review 1
Adaptive weight factor estimation from user review 1
 
Integrated institutional portal
Integrated institutional portalIntegrated institutional portal
Integrated institutional portal
 
Embassy
EmbassyEmbassy
Embassy
 
Crm
Crm Crm
Crm
 
Mutual distance bounding protocols
Mutual distance bounding protocolsMutual distance bounding protocols
Mutual distance bounding protocols
 
Moderated group authoring system for campus wide workgroups
Moderated group authoring system for campus wide workgroupsModerated group authoring system for campus wide workgroups
Moderated group authoring system for campus wide workgroups
 
Efficient spread spectrum communication without pre shared secrets
Efficient spread spectrum communication without pre shared secretsEfficient spread spectrum communication without pre shared secrets
Efficient spread spectrum communication without pre shared secrets
 
Comments on “mabs multicast authentication based on batch signature”
Comments on “mabs multicast authentication based on batch signature”Comments on “mabs multicast authentication based on batch signature”
Comments on “mabs multicast authentication based on batch signature”
 
Energy-Efficient Protocol for Deterministic and Probabilistic Coverage In Sen...
Energy-Efficient Protocol for Deterministic and Probabilistic Coverage In Sen...Energy-Efficient Protocol for Deterministic and Probabilistic Coverage In Sen...
Energy-Efficient Protocol for Deterministic and Probabilistic Coverage In Sen...
 
Energy efficient protocol for deterministic
Energy efficient protocol for deterministicEnergy efficient protocol for deterministic
Energy efficient protocol for deterministic
 
A Privacy-Preserving Location Monitoring System for Wireless Sensor Networks
A Privacy-Preserving Location Monitoring System for Wireless Sensor NetworksA Privacy-Preserving Location Monitoring System for Wireless Sensor Networks
A Privacy-Preserving Location Monitoring System for Wireless Sensor Networks
 

Recently uploaded

1029 - Danh muc Sach Giao Khoa 10 . pdf
1029 -  Danh muc Sach Giao Khoa 10 . pdf1029 -  Danh muc Sach Giao Khoa 10 . pdf
1029 - Danh muc Sach Giao Khoa 10 . pdf
QucHHunhnh
 
BASLIQ CURRENT LOOKBOOK LOOKBOOK(1) (1).pdf
BASLIQ CURRENT LOOKBOOK  LOOKBOOK(1) (1).pdfBASLIQ CURRENT LOOKBOOK  LOOKBOOK(1) (1).pdf
BASLIQ CURRENT LOOKBOOK LOOKBOOK(1) (1).pdf
SoniaTolstoy
 

Recently uploaded (20)

The Most Excellent Way | 1 Corinthians 13
The Most Excellent Way | 1 Corinthians 13The Most Excellent Way | 1 Corinthians 13
The Most Excellent Way | 1 Corinthians 13
 
Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"
Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"
Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"
 
9548086042 for call girls in Indira Nagar with room service
9548086042  for call girls in Indira Nagar  with room service9548086042  for call girls in Indira Nagar  with room service
9548086042 for call girls in Indira Nagar with room service
 
microwave assisted reaction. General introduction
microwave assisted reaction. General introductionmicrowave assisted reaction. General introduction
microwave assisted reaction. General introduction
 
Class 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdfClass 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdf
 
Mattingly "AI & Prompt Design: The Basics of Prompt Design"
Mattingly "AI & Prompt Design: The Basics of Prompt Design"Mattingly "AI & Prompt Design: The Basics of Prompt Design"
Mattingly "AI & Prompt Design: The Basics of Prompt Design"
 
Interactive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communicationInteractive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communication
 
A Critique of the Proposed National Education Policy Reform
A Critique of the Proposed National Education Policy ReformA Critique of the Proposed National Education Policy Reform
A Critique of the Proposed National Education Policy Reform
 
Grant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy ConsultingGrant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy Consulting
 
Advance Mobile Application Development class 07
Advance Mobile Application Development class 07Advance Mobile Application Development class 07
Advance Mobile Application Development class 07
 
Nutritional Needs Presentation - HLTH 104
Nutritional Needs Presentation - HLTH 104Nutritional Needs Presentation - HLTH 104
Nutritional Needs Presentation - HLTH 104
 
Sanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdfSanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdf
 
1029 - Danh muc Sach Giao Khoa 10 . pdf
1029 -  Danh muc Sach Giao Khoa 10 . pdf1029 -  Danh muc Sach Giao Khoa 10 . pdf
1029 - Danh muc Sach Giao Khoa 10 . pdf
 
IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...
IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...
IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...
 
Unit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptxUnit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptx
 
BASLIQ CURRENT LOOKBOOK LOOKBOOK(1) (1).pdf
BASLIQ CURRENT LOOKBOOK  LOOKBOOK(1) (1).pdfBASLIQ CURRENT LOOKBOOK  LOOKBOOK(1) (1).pdf
BASLIQ CURRENT LOOKBOOK LOOKBOOK(1) (1).pdf
 
fourth grading exam for kindergarten in writing
fourth grading exam for kindergarten in writingfourth grading exam for kindergarten in writing
fourth grading exam for kindergarten in writing
 
General AI for Medical Educators April 2024
General AI for Medical Educators April 2024General AI for Medical Educators April 2024
General AI for Medical Educators April 2024
 
social pharmacy d-pharm 1st year by Pragati K. Mahajan
social pharmacy d-pharm 1st year by Pragati K. Mahajansocial pharmacy d-pharm 1st year by Pragati K. Mahajan
social pharmacy d-pharm 1st year by Pragati K. Mahajan
 
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
 

Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes

  • 1. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. IEEE TRANSACTIONS ON MOBILE COMPUTING 1 Estimating Parameters of Multiple Heterogeneous Target Objects Using Composite Sensor Nodes Hiroshi Saito, Fellow, IEEE, Shinsuke Shimogawa, Sadaharu Tanaka, and Shigeo Shioda, Member, IEEE Abstract—We propose a method for estimating parameters of multiple target objects by using networked binary sensors whose locations are unknown. These target objects may have different parameters, such as size and perimeter length. Each sensors, which is incapable of monitoring the target object’s parameters, sends only binary data describing whether or not it detects target objects coming into, moving around, or leaving the sensing area at every moment. We previously developed a parameter estimation method for a single target object. However, a straight-forward extension of this method is not applicable for estimating multiple heterogeneous target objects. This is because a networked binary sensor at an unknown location cannot provide information that distinguishes individual target objects, but it can provide information on the total perimeter length and size of multiple target objects. Therefore, we propose composite sensor nodes with multiple sensors in a predetermined layout for obtaining additional information for estimating the parameter of each target object. As an example of a composite sensor node, we consider a two-sensor composite sensor node, which consists of two sensors, one at each of the two end points of a line segment of known length. For the two-sensor composite sensor node, measures are derived such as the two sensors detecting target objects. These derived measures are the basis for identifying the shape of each target object among a given set of categories (for example, disks and rectangles) and estimating parameters such as the radius and lengths of two sides of each target object. Numerical examples demonstrate that networked composite sensor nodes consisting of two binary sensors enable us to estimate the parameters of target objects. Index Terms—Sensor network, estimation, target object, composite sensor node, ubiquitous network, integral geometry, geometric probability. ✦ 1 I NTRODUCTION extract significant information. This sensing paradigm is appropriate for the described situation. Small and low-cost sensors can be used to build wireless Unfortunately, we have not determined a theory that sensor networks [1], [2], [3]. They have communication enables us to implement this paradigm. As a typical capabilities and built-in batteries to communicate via a and challenging application, we investigated a problem wireless link for transmitting sensed data of detected of estimating parameters related to the shape and size events of interest. A newly proposed development, such of target objects moving in a monitored area by using as a wide area ubiquitous network [4], supports sensors many simple randomly distributed sensors. We address with low performance and functionalities and a long- this problem and present a theory for estimating the range, low-speed wireless link with very low power con- parameters of multiple target objects by using networked sumption. These sensors can be implemented as single binary sensors whose locations are unknown. An in- chips, lowering their cost. As a result, we can deploy dividual sensor is simple. It monitors its environment such sensors like the scattering of dust [5]. Because there and reports whether or not it detects a target object. It are many sensors, we cannot carefully design the loca- does not have a positioning function, such as a GPS, or tion of each one. GPS cannot be used because the sensors functions for monitoring the target object size and shape, should have limited capability and power consumption. such as a camera, and it can be placed without careful This situation leads us to a new sensing paradigm. design. However, by collecting reports from individual That is, instead of having a few sensors with advanced sensors, we can statistically estimate the parameters of functions and high performance, many sensors with target objects. simple functions and low performance are distributed In addition to theoretical contributions, this work is randomly. They are networked and send reports, each of useful to some application scenarios that require low which includes only an insignificant amount of informa- cost, a wide coverage, and low power consumption tion, over a wireless link. However, as a whole, we can and accept rough estimation of the size and shape of target objects. For example, we may need to identify the ¯ H. Saito and S. Shimogawa are with NTT Service Integration Laboratories, 3-9-11, Midori-cho, Musashino-shi, Tokyo, 180-8585, Japan. number and kinds of animals and vehicles in a large E-mail: saito.hiroshi@lab.ntt.co.jp area to observe the wildlife, prevent poaching, or limit ¯ S. Tanaka and S. Shioda are with Chiba University. the number of tourist vehicles. We can easily obtain bi- nary sensors of sound volume, acceleration, infrared, or Digital Object Indentifier 10.1109/TMC.2011.65 1536-1233/11/$26.00 © 2011 IEEE
  • 2. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. IEEE TRANSACTIONS ON MOBILE COMPUTING 2 magnetic sensors for such scenarios. These sensors with for such a purpose. In particular, [8] and [14] directly normal detection ranges are not expensive. Therefore, we apply the results of integral geometry discussed in can provide a large number of these sensors to cover a Chapter 5 and Section 6.7 in [15] to the analysis of large area at a reasonable cost. By distributing these bi- detecting an object moving in a straight line and to the nary sensors without expensive power-consuming GPSs evaluation of the probability of -coverage. In addition, or time-consuming carefully-designed placement in a [16] applies integral geometry to the analysis of straight large area, our proposed method can work. line routing, which is an approximation of the shortest This research is an extension of our prior study [6], path routing, and [17] uses it to operate sensors in an which was based on the coverage process theory [7] energy-conserving way. and its application to sensor networks [8], [9], [10]. This The rest of this paper is organized as follows. study emphasized the robustness of the derived formula Section 2 describes a basic model and previous results for a number of sensors detecting a target object. In for the model. In Section 3, we introduce composite [6], sensors that measure the size of a detected part sensor nodes and describe an extended model (static of a target object are used for estimating the overall submodel). We discuss the analysis for deriving mea- size. However, in [11], we developed a shape and size sures in Section 4. In Section 5, we discuss the dynamic estimation method using binary sensors for sending submodel of the extended submodel. We then propose reports on whether or not they detect the target object. an estimation method that uses composite sensor nodes That study assumed that both the target object and the for target objects with different parameters in Section 6. sensing area are convex or that the sensing area is disk- We show numerical examples in Section 7, and conclude shaped. This developed method [11] was evaluated in this paper in Section 8. an experiment where the target object was a box and the sensor was an infrared distance measurement sensor [12]. We also extended the estimation method in [11] to 2 P REVIOUS WORK apply to non-convex target objects [13]. 2.1 Basic model These methods we developed in [11], [13] did not take into account when multiple target objects are in the To describe the results of the previous work, which is monitored area. This work addresses such a case, i.e., the basis of the current work, we discuss the basic model multiple target objects that may have different shapes that the previous work used. and sizes coming into, moving around, and leaving the A sensor network operator deploys sensors in a con- area monitored by sensors. Each sensor, whose location vex area ¨ in 2-dimensional space ʾ , but the operator is unknown, sends a report on whether or not at least does not know the sensor locations. Each sensor has one target object is detected within its sensing area. The a sensing area, monitors the environment, and detects sensor network operator collects these reports to identify events within that area. (This model is called the Boolean the shape and to estimate parameters, such as sizes sensing model [9], [18], [19], [20] because it is clear and perimeter lengths, of the target objects. However, whether a point is within a sensing area or not.) A sensor identifying the shape and estimating the parameters of sends a detection report if a target object is in its sensing multiple heterogeneous (i.e., having different parame- area, and it sends a no-detection report if there is no ters) target objects is difficult, and a straight-forward target object in that area. (It is possible to assume that a extension of the estimation method we previously de- sensor does not send a report if there is no target object veloped [11] is not applicable. Thus, we introduce the in that area and that the sensor network operator judges concept of composite sensor nodes where multiple sim- no-detection if there is no report within a timeout. This ple binary sensors are arranged in a predetermined scenario can reduce the power consumption of sensors, layout, such as a line. Sensors in a composite sensor but the sensor network operator cannot distinguish no- node provide local and relative information even if the detection from the loss of reports. For simplicity, this location of that node is unknown. This is because the paper assumes that no-report is sent if there is no target relative locations of these sensors are predetermined. In object in that area.) this sense, the composite sensor node is an intermediate Assume that the -th sensor is located at ´Ü Ý µ and concept between a simple and an advanced sensor node the sensing area is rotated by from the referenced equipped with GPS. With these sensors, we can estimate position. Let ´Ü Ý µ ʾ be the sensing area the parameters of multiple heterogeneous target objects. where ½ ¾ . The -th sensor has communication The estimation method using composite sensor nodes capability and can send a report Á . Here, Á is 1 if is derived from integral geometry and geometric prob- it detects the target object and 0 if otherwise. That is, ability, which are useful tools for analyzing a geometric Á ½´ ´Ü Ý µ Ì µ, where ½´ µ is an indicator or spatial structure. Ubiquitous access networks, such function that becomes 1 if a statement is true and 0 if as wireless sensor networks, require analysis of the otherwise. At Ø Ø , the -th sensor sends a report, Á ´Ø µ, geometric or spatial structure of an object, and there is describing whether or not it detects the target object. The literature (including one of our papers [11]) describing sensor network operator receives the report from each the use of integral geometry and geometric probability sensor through the network.
  • 3. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. IEEE TRANSACTIONS ON MOBILE COMPUTING 3 Sensors are classified into multiple types according to 3 C OMPOSITE SENSOR NODE their sensing areas. That is, different types of sensors Estimating the unknown parameters of multiple target have different sensing area sizes or shapes. (A different objects is difficult. Even when the number of target sensing area can often be implemented by changing the objects is known, a straight-forward extension of the sensitivity parameter of a sensor.) Type- sensors, of estimation method for a single target object to one for which sensing area is denoted by , are deployed with multiple target objects is not applicable if the parameters mean density . Assume that the first to Ò½ -th sensors of each target object are different. To estimate the pa- are type-1 sensors, the ´Ò½ · ½µ-th to Ò¾ -th sensors are rameters of each target object, we introduce a composite type-2 sensors, ... In general, Ò  ½ · ½ ¡ ¡ ¡ Ò -th sensors are type- sensors. Á ØÑ Ø Ø½ Ò È È Ò  ½ ·½ Á ´Øµ Ñ de- sensor node, which consists of multiple sensors arranged in a predetermined layout, such as a line. Instead of notes the time average of the number of type- sensors deploying individual sensors, the sensor network op- detecting the target object at each measurement epoch. erator deploys composite sensor nodes. Although the We assume that the sensor network operator knows the locations of the composite sensor nodes are unknown, sensor type of each sensor sending a report. local and relative information among multiple sensors Consider a target object Ì in 2-dimensional space ʾ . in a predetermined layout of a composite sensor node Its size is Ì , and its perimeter length is Ì . (In the becomes available because the relative locations of these remainder of this paper, Ü denotes the size of Ü and sensors are known. Ü denotes the perimeter length of Ü.) Define a detectable area such that sensors will 3.1 Straight-forward extension to multiple target ob- detect the target object if and only if they are located jects in a detectable area with a certain rotation angle. That is, ½´ ´Ü Ý µ Ì µ ½ if and only if Before introducing composite sensor nodes, we explain ´Ü Ý µ ¾ an extension of Eq. (1) to an equation applicable to . Equivalently, ´Ü Ý µ ´Ü Ý µ Ì ÊÊÊ . The detectable area size is defined as multiple target objects. Proposition 1: Assume that there are Ò Ì convex and ÜÝ ¾ Ü Ý ¾ . We should note that and Ì . When bounded target objects and that the sensing area is depends on both is disk-shaped, also bounded and convex. Let be the detectable areas of the -th target object and £ does not depend on . When does not depend on , ´Ü Ý µ ´Ü Ý µ Ì . If we define for simplicity. for any , Ì Ò Ò Ì 2.2 Previous results for basic model £ ½ ´ Ì µ¡ · Ì · ÒÌ (5) ¾ ½ ½ In this section, we summarize our previous results, which are used in the current work. For the convex target This is because È where Ì denotes the -th target object ´½ and because ÒÌ µ. £ object, we have obtained the following equation, and a shape and size estimation method was developed based satisfies Eq. (1). Therefore, we can estimate ÒÌ ½ Ì and È ÒÌ ½ Ì È on this equation [11]: Assume that both the target object Ì and the sensing area are bounded and convex. Then, instead of Ì and Ì in Eqs. (3) and (4), respectively. Consequently, if we can assume that all the target objects ½ ¾ Ì ¡ · Ì · (1) have the same shape and size, we can estimate the size and shape (size and perimeter length) of each target object. However, if not, we cannot estimate the size and Based on the fact that Á , we developed shape (size and perimeter length) of each target object the following estimation method. (1) At Ø , receive the re- È È no matter how many types of sensors we introduce. port Á ´Ø µ from each sensor whose location is unknown. Ñ Ò (2) Calculate Á ½ Ò  ½ ·½ Á ´Ø µ Ñ for each sensor type ( ½ ¾). (3) For two unknown parameters 3.2 Extended model (static submodel) Ì and Ì , use Eq. (1) with the estimator Á of For analyzing an estimation method for multiple hetero- and solve the following equations for ½ ¾. geneous target objects, we introduce a new model. A sensor network operator deploys composite sensor ´ ½ ¾ Ì ¡ · Ì · µ Á (2) nodes in a convex area ¨ in 2-dimensional space ʾ , but the operator does not know their locations. A composite where Ì Ì are estimators of Ì Ì . That is, sensor node consists of multiple sensors arranged in a predetermined layout. We now consider a two-sensor Ì ¾ ´Á½ ½   Á¾ ¾  ½ · ¾ µ composite sensor node, i.e., a composite sensor node ½   ¾ (3) with two sensors. These two sensors are the same as those described in the basic model. That is, they are Ì ¾ ´ ½   Á½ ½µ · ½ ´ ¾   Á¾ ¾µ simple binary sensors. For simplicity, in the remainder ¾   ½ (4) of this paper, we assume that the sensing area of each
  • 4. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. IEEE TRANSACTIONS ON MOBILE COMPUTING 4 individual sensor in a composite sensor node is disk- 4 A NALYSIS FOR EXTENDED MODEL ( STATIC shaped. SUBMODEL ) 4.1 General results There are multiple types of composite sensor nodes. Type- composite sensor nodes are randomly deployed In this subsection, we provide general measure proposi- with mean density ¼. Each of the two sensors in a tions that only one (both) of the sensors in a composite type- composite sensor node has a disk-shaped sensing sensor node detects a target object. Let ѽ ´Ì µ be the area with radius Ö and is located at each of the two end measure of a set of composite sensor nodes in which one points of a line segment of length Ð . Assume that the first of the two sensors in each composite sensor node detects to Ò½ -th composite sensor nodes are type-1 composite a target object Ì and the other does not. (According to sensor nodes and the ´Ò½ · ½µ-th to Ò¾ -th composite ѽ ´Ì µ Ê Appendix A, ѽ ´Ì µ can be written by an integral form: Ø.) Roughly, ѽ ´Ì µ is a sensor nodes are type-2 composite sensor nodes, ... In ´Ü½ ܾ µ¾ × Ô general, Ò  ½ · ½ ¡ ¡ ¡ Ò -th composite sensor nodes are non-normalized probability that one of the two sensors type- composite sensor nodes for ½ ¡ ¡ ¡  , where in a composite sensor node detects a target object Ì and  is the number of composite sensor node types. (For the other does not, where the normalizing constant is Ñ ½ ¾, Ð ½ Ð ¾ or Ö ½ Ö ¾ .) (the measure of a set of composite sensor nodes placed Let Á be the report of the -th sensor of the -th A. Similarly, let Ѿ ´Ì µ Ê in (included in or intersects with) ¨), given in Appendix ´Ü½ ܾ µ¾ Ô Ø be the composite sensor node for ½ ¾, and Á is 1 if it measure of a set of composite sensor nodes in which È detects the target object and 0 if otherwise. Let Á¾ Ò Ò  ½ ·½ ½´Á½ Á¾ ½µ be the number of type- both sensors of each composite sensor node detect a target object Ì . Ѿ ´Ì µ is a non-normalized probability composite sensor nodes in each of which two sensors that both of the two sensors in a composite sensor node Ò È detect at least one target object at a single measurement detect a target object Ì . (When we need to indicate epoch. Similarly, define Á½ Ò  ½ ·½ ½´ Á½ ´½   the parameters Ð Ö of the composite sensor node to Á¾ µ · ´½   Á½ µÁ¾ ½µ, which denotes the number evaluate the measures, we use the notations ѽ ´Ì Ð Öµ of type- composite sensor nodes in each of which one and Ѿ ´Ì Ð Öµ. When we need to indicate the parameters of two sensors detects at least one target object at a single of the -th target object in Ѿ for the type- composite measurement epoch. sensor node, we use the notation Ѿ ´¢ Ð Ö µ instead of Ѿ ´Ì Ð Ö µ, where ¢ is defined later.) There are ÒÌ target objects in 2-dimensional space Proposition 2: Let ѽ ´ µ be the measure of a set of type- ʾ , where ÒÌ is known. Let Ì be the -th target object composite sensor nodes in which only one of the two (½ ÒÌ ). (We propose the stochastic geometric filter, sensors in each composite sensor node detects any target which can estimate the number of target objects [21].) In object, and let Ѿ ´ µ be the measure of a set of type- this static submodel, target objects do not move. composite sensor nodes in which both sensors of each composite sensor node detect any target object. If there For type- composite sensor nodes, define a is no overlap between composite-detectable areas ½ composite-detectable area such that the sensors of for any ½ ¾ (½ ½ ¾ ÒÌ ), ¾ the composite sensor nodes will detect the -th target ÒÌ object if and only if they are located in a composite- ѽ ´ µ ѽ ´Ì Ð Ö µ (6) detectable area Ê . That is, when the locations ½ of the two sensors of a composite sensor node at the ÒÌ two end points of a line segment of length Ð are Ü and Ѿ ´ µ Ѿ ´Ì Ð Ö µ (7) their disk-shaped sensing areas are ´Ü µ ( ½ ¾), ½ ½´´ ´Ü½ µ Ì µ ´ ´Ü¾ µ Ì µ µ ½ if and only if £ ´Ü½ ܾ µ ¾ , where ܽ   ܾ ¾ о . Similarly, we This is because, if there is no overlap of detectable × define a single detectable area (a double detectable areas, then the event that one of the two sensors (both area ) such that only one (both) of the two sensors of the two sensors) in a composite sensor node detects of type- composite sensor nodes will detect the -th Ì or Ì is the sum of the two events. One is the event target object if and only if they are located in a single that one of the two sensors (both of the two sensors) in detectable area × Ê (in a double detectable the composite sensor node detects Ì , and the other is area Ê ). That is, when the locations of the the event that one of the two sensors (both of the two two sensors of a composite sensor node at the two sensors) in the composite sensor node detects Ì . end points of a line segment of length Ð are Ü and We can derive other measures on this basis. For ex- their disk-shaped sensing areas are ´Ü µ ( ½ ¾), ample, the measure for at least one sensor in a type- ½´ ´Ü½ µ Ì µ ¡ ½´ ´Ü¾ µ Ì µ ½ if and only composite sensor node detecting a target object is if ´Ü½ ܾ µ ¾ , where ܽ   ܾ ¾ о . In addition, ѽ ´ µ · Ѿ ´ µ. ×   . (We may remove in , × , Remark 1: If there are overlaps of detectable areas of to simplify the notation, when are not specified.) individual target objects (i.e., if ½ ¾ ), the
  • 5. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. IEEE TRANSACTIONS ON MOBILE COMPUTING 5 estimation error may increase (see Section 7. Numerical 4.2.1 Definitions and notations examples). To avoid overlap, small sensing areas are Here, we provide a list of definitions and notations used preferable. In a large sensing area, it may not be able in this subsection. to detect a small gap between two target objects. This is ¯ ¢ : a vector of parameters describing the -th target similar to the phenomenon of a large-sensing-area sensor object. not detecting a small hole or a small concave part of a ¯ ÒÔ ´ µ: the number of parameters in ¢ . (For simplic- target object, causing a large estimation error [11], [13]. ity, ÒÔ ´ µ is constant in the following if not explicitly For no overlaps between composite-detectable areas, indicated.) two conditions are required. The first is that a single sensor in a composite sensor node should not simulta- ¯ ¢´ µ ´¢ ¡ ¡ ¡ ¢ µ. neously detect multiple target objects. The second is that ¯ ¢ ¢´½ ÒÌ µ. the two sensors in a composite sensor node should not ¯ Û ´Ð Ö µ. simultaneously detect multiple target objects. The first ¯ Ï Û ÈÛ ´ ½ ¡ ¡ ¡  µ. condition is identical to that in which the detectable areas of individual target objects for a simple sensor Ú Û ¯ ´¢ µ ÒÌ ½ Ѿ ´¢ µ. Û do not overlap, and is also required in Proposition 1, Ú Ï ¯ ´¢ µ Ú Û Ú Û ´ ´¢ ½ µ ¡ ¡ ¡ ´¢  µµ. which does not use a composite sensor node. However, the second condition is not needed when we do not use 4.3 Definition and proposition of observability the composite sensor node. This can be a new cause of errors, although composite sensor nodes can obtain new Definition 1: A value vector ¢ of parameter vector ¢ is information. To reduce this error, a shorter Ð is better. observable if there exists a set of composite sensor node Both conditions can be easily satisfied when the target parameter values Ï ´Ð½ Ö½ µ ¡ ¡ ¡ ´Ð Ö µ satisfying object’s density is low. £ Ú Ï that ´¢ µ Ú ´¢¼ Ï µ for any ¢¼ ¢ ¢¼ ¾ ËÔ for a The following proposition means that the expected given feasible parameter space Ë Ô . Here, Ï is called an number of type- composite sensor nodes in which only observing parameter set. £ one (both) of the two sensors detects a target object is Under an ideal situation (that is, there are no proportional to ѽ ´ µ (Ѿ ´ µ). This is natural because approximation or measurement errors), ´¢ µ Ú Ï of the definition of ѽ ´ µ (Ѿ ´ µ). See Appendix A for ´Á¾ ½ ¡ ¡ ¡ Á¾  µ. Therefore, roughly speaking, Definition mathematical details. In addition, the following propo- 1 implies that if obtained sensor reports can uniquely sition is valid for any shaped target object. determine ¢ under an ideal situation when we use a Proposition 3: Let Æ ´½ µ be the number of type- certain set of composite sensor node parameter values composite sensor nodes in which one of the two sensors Ï , ¢ is said to be “observable.” The statement that detects a target object, and let Æ ´¾ µ be the number obtained sensor reports can uniquely determine ¢ means of type- composite sensor nodes in each of which two that any other value vector ¢¼ of parameter vector ¢ is sensors detect a target object. If the composite sensor not consistent with the obtained sensor reports. nodes are distributed in a sufficiently large area, The definition of observability requires the uniqueness of the parameter value vector that is consistent to sensor Æ ´½ µ ѽ ´ µ (8) reports. However, it does not require the uniqueness of Æ ´¾ µ Ѿ ´ µ (9) Ú Ï ´¢ µ over the entire domain of ¢. In addition, an observing parameter set can depend on ¢. £ Thus, the following proposition is directly derived Precisely, Eqs. (8) and (9) are affected by the shape from the definition of observability. of ¨, the sensor-deployed area (Appendix A). However, Ú Ï Proposition 4: If ´¢ µ is given and if ¢ is observable if the border effect (the number of composite sensor with an observing parameter set Ï , we can uniquely nodes intersecting the border of ¨) is small, they are and exactly estimate ¢. £ independent of the shape of ¨. Practically, this is the Because ´¢ µ Ú Ï ´Á¾ ½ ¡ ¡ ¡ Á¾  µ if there is no case. approximation or measurement errors, we can uniquely Note that the sample of the random variable Æ ´½ µ and exactly estimate observable parameter values by is Á½ and that of Æ ´¾ µ is Á¾ , and that Æ ´½ µ using an observing parameter set Ï and sensor reports. Á½ and Æ ´¾ µ Á¾ . Equivalently, if ¢ is observable, there exists an observing parameter set, and we can uniquely and exactly estimate 4.2 Observability ¢ by using it and sensor reports if there is no approxi- mation or measurement errors. On the other hand, even In the remainder of this section, target objects are as- for observable parameter values, if the parameters of sumed to be convex. Without loss of generality, we can composite sensor nodes are not appropriately chosen, assume that ½ ¡ ¡ ¡ ÒÌ and н   ¾Ö½ ¡ ¡ ¡ Р  ¾Ö . we may not be able to appropriately estimate them. Here, Ô is the diameter of the -th target object, that is Unfortunately, we cannot judge whether a given is Ï Ñ Ü´Ü½ ݽ µ ´Ü¾ ݾ µ¾Ì ´Ü½   ܾ µ¾ · ´Ý½   ݾ µ¾ . an observing parameter set without knowing values of
  • 6. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. IEEE TRANSACTIONS ON MOBILE COMPUTING 6 Ï ¢ or cannot provide , which is an observing parameter 4.4 Derivation of measures set for any values of ¢. In the following subsections, we derive the measures If ½ ¡¡¡ ÒÌ , we can get a simplified sufficient ѽ and Ѿ for a certain class of target objects (disk- condition for observability. This condition can be deter- shaped and rectangular target objects) as examples. mined by an individual target object. (Consequently, through Eqs. (6), (7), (8), and (9), Ñ ½ and Ѿ , Æ ´½ µ and Æ ´¾ µ can be obtained.) We first Ï Lemma 1: If there exists a set of composite sensor node parameter values ´Ð½ Ö½ µ ¡ ¡ ¡ ´ÐÒÔ ÒÌ ÖÒÔ ÒÌ µ derive the measures ѽ and Ѿ for a disk-shaped target satisfying that  ½ д  ½µÒÔ ·½   ¾Ö´  ½µÒÔ ·½ ¡¡¡ object. Second, we derive measures for rectangular target д  ½µÒÔ ·ÒÔ   ¾Ö´  ½µÒÔ ·ÒÔ for ½ Û Û ÒÌ and that ´Ñ¾ ´¢¼ ´  ½µÒÔ ·½ µ ¡ ¡ ¡ Ѿ ´¢¼ ´  ½µÒÔ ·ÒÔ µµ objects. Finally, they are derived when there are disk- Û Û ´Ñ¾ ´¢ ´  ½µÒÔ ·½ µ ¡ ¡ ¡ Ѿ ´¢ ´  ½µÒÔ ·ÒÔ µµ when ¢¼ shaped and rectangular target objects. ¢ ¢¼ ¾ ËÔ ´ µ for a given feasible parameter space ËÔ ´ µ 4.4.1 Disk-shaped target objects for ½ ÒÌ , ¢ is observable with an observing parameter set Ï .£ Proof: Assume that ¢¼ £ ¢ £ and ¢¼ ¢ for £ When a target object is disk-shaped, we can obtain explicit formulas. We derive ѽ and Ѿ under the as- sumption that there is a single target object whose radius ÒÌ . is Ê, the radius of each sensing area in a composite Suppose that  ½ д  ½µÒÔ ·½   ¾Ö´  ½µÒÔ ·½ ¡¡¡ sensor node is Ö, and the distance between the two д  ½µÒÔ ·ÒÔ   ¾Ö´  ½µÒÔ ·ÒÔ for ½ ÒÌ . sensors in the node is Ð. Consider the type- composite sensor nodes where ´ £   ½µÒÔ · ½ ´ £   ½µÒÔ · ÒÔ . Note that, if From Appendix B, Û È £ , Ѿ ´¢ µ ¼Ò because Ð   ¾Ö . There- ѽ ´Ê Ð Öµ È Ú Û fore, ´¢ Ú Û µ Ì Ñ ´¢ µ and ´¢¼ £ ¾ ¼ µ for ´ £  ½µÒÔ ·½ µ ¾ ¾´Ê · Öµ¾ × Ò ½ ´ ¾´Êзֵ µ Ô ÒÌ £ Ѿ ´¢ ´ £  ½µÒÔ ·ÒÔ . ·Ð ´Ê · Öµ¾   о ¾´Ê · Öµ Ú Û Ú Û Hence, ´¢ µ   ´¢¼ µ ´Ñ¾ ´¢ £ µ   Ѿ ´¢¼ £ µµ. ¾ ¾ ´Ê · Öµ¾ for for ¾´Ê · Öµ Ð Ð. Ú Û Ú Û Ú Û Ú Û According to the assumption of this lemma, for (10) ¢¼ £ ¢ £ , ´ ´¢ ´ £  ½µÒÔ ·½ µ ¡ ¡ ¡ ´¢ ´ £  ½µÒÔ ·ÒÔ µµ Ѿ ´Ê Ð Öµ ´ ´¢¼ ´ £  ½µÒÔ ·½ µ ¡ ¡ ¡ ´¢¼ ´ £  ½µÒÔ ·ÒÔ µµ. ´Ê Ð Ö µ for ¾´Ê · Öµ Ú Ï Ú Ï Ð, Consequently, ´¢ µ ´¢¼ µ if ¢¼ ¢. £ ¼ otherwise, (11) In practice, we are likely to face the following situ- ation: The target object shape can be categorized into where ´Ê Ð Öµ Ô ´Ê · Öµ¾ ´   ¾ × Ò ½ ´ ¾´Êзֵ µµ   several categories, such as disks and rectangles, and Ð ´ Ê · Ö µ¾ Ð ¾  . we may not know how many target objects belong to Remark 2: When there are Ò Ì disk-shaped target objects each category. Note that ÒÔ ´ µ is likely to depend on the Ï and the radius Ê of the -th target object satisfies category to which the -th target object belongs. Let be ʽ ¡¡¡ ÊÒÌ , that satisfies н   ¾Ö½ ¾Ê½ the number of target objects in the -th category and Ò Ð¾   ¾Ö¾ ¾Ê¾ ¡ ¡ ¡ ÐÒÌ   ¾ÖÒÌ ¾ÊÒÌ is an observing the number of categories. ½ ¡ ¡ ¡ Ò are also unknown parameter set, due to Lemma 1. This is because ´Ê Ð Öµ parameters. Similar to Proposition 4, the following corol- is an increasing function of Ê. £ lary shows that we can estimate ½ ¡ ¡ ¡ Ò as well as In the remainder of this paper, if we need to explicitly other observable parameters ¢. Ú Ï indicate “disk-shaped target object” for these measures Corollary 1: If ´¢ µ is given and if ¢ and values of ѽ and Ѿ , we use the notations ѽ and Ѿ . ½ ¡ ¡ ¡ Ò are observable, we can uniquely and exactly estimate them. £ 4.4.2 Rectangular target objects Proposition 4, Lemma 1, and the corollary mentioned above mean that if we can provide more than È ÒÔ ´ µ This subsection analyzes rectangular target objects. Con- sider a single rectangular target object with two sides types of composite sensor nodes with appropriate Ð Ö and a single type of composite sensor node whose sen- and a sufficiently large number of samples of sensed re- sors’ sensing-area radius is Ö and where the distance be- sults, we can estimate observable values of parameters of tween the sensors is Ð. The necessary and sufficient con- any convex target object by using two-sensor composite dition of the first (second) sensor in a composite sensor sensor nodes. To concretely obtain estimates, a calcu- node detecting the target object is that the location of the lation method for Ѿ ´¢ Ð Öµ is required. As examples, first (second) sensor is in . Here, is the detectable area we provide formulas to calculate Ѿ ´¢ Ð Öµ for a certain of this rectangular target object when a basic (i.e., non- class of target objects. Theoretically, a simulation is ap- composite) disk-shaped sensing area with a radius Ö is plicable by doing a simulation for various values of ¢ used. That is, ´Ü ݵ Ñ Ò  ¾ ܼ ¾   ¾ ݼ ¾ ´Ü  for each pair of ´Ð Ö µ to obtain Ѿ ´¢ Ð Ö µ. However, ܼ µ¾ · ´Ý   Ý ¼ µ¾ Ö¾ . To simplify the calculation, we practically, the applicability of the simulation is limited introduce ´Ü ݵ   ¾   Ö Ü ¾· Ö   ¾   Ö to special cases, for example, those in which the shapes Ý ¾ · Ö instead of (Fig. 1). That is, . of the target objects and the ranges of parameter values Then, the necessary and sufficient condition of the first are roughly known in advance. (second) sensor in a composite sensor node detecting the
  • 7. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. IEEE TRANSACTIONS ON MOBILE COMPUTING 7 of ´Ð Ö µ, which satisfies Ð   ¾Ö Ñ Ò´ µ is not D Ï D included in an observing parameter set. On the other hand, if ¾ · ¾ Ô ½ ½ ¡¡¡ ¾ · ¾ , ÒÌ ÒÌ that satisfies Ñ Ü´ ´  ½ · ¾Ö¾ µ ¾·´  ½ · ¾Ö¾ µ¾ · Ô · ·¾ l ¾Ö¾ · ¾Ö¾ µ о  ½ о ¾ ¾ Ö¾ Ö¾ t p H r Ö¾  ½ о о  ½ · Æ is an observing parameter set where Æ is a sufficiently small positive scalar. See Appendix D b for details. £ a In the remainder of this paper, if we need to explicitly indicate “rectangle” for these measures Ñ ½ and Ѿ , we use the notations ѽ Ö and Ѿ Ö . r G 4.4.3 Combinations of disk-shaped and rectangular tar- get objects Let Ò be the number of disk-shaped target objects and Fig. 1. Analysis of two-sensor composite sensor node for ÒÖ ÒÌ   Ò be the number of rectangular target objects. rectangular target object Ò and ÒÖ are unknown parameters. As the measures are additive if ½ ¾ for any ½ ¾ (½ ½ ¾ ÒÌ ½ ¾ ), we can easily obtain target object is approximately equivalent to the location Ò of the first (second) sensor being in . We use this ѽ ´µ ѽ ´Ê Ð Ö µ approximation and derive measures. Because brute-force ½ ÒÖ but lengthy computations are needed, we show only the results here. The computation details are in Appendix C. · ѽ Ö ´ Ð Ö µ (14) Define · ¾Ö , · ¾Ö, « Ñ Ò´ µ, and Ò ½ ¬ Ñ Ü´ µ. Ѿ ´ µ Ѿ ´Ê Ð Ö µ ½ ѽ ´ Ð Öµ ÒÖ Ð´ · µ ¾Ð ¾   for Ð «, · Ѿ Ö ´ Ð Ö µ (15) «¬ Ó×  ½ ´« е · Ь ½ ¬ о «¾ · «¾ Ô     for « Ð ¬ , where Ê is the radius of the -th disk-shaped target ob- ´Ô  ½ ´ Ð µ · Ó× ½ ´ Ð µµ Ó× (12) ject and are the side lengths of the -th rectangular Ô¾ Ô   о   ¾   о   ¾ target object. ·¾´ · · µ ¾ ¾ о Ô ¾Ð for ¬ · ¾, ¾ for · ¾ Ð, 5 E XTENDED MODEL ( DYNAMIC SUBMODEL ) 5.1 Model description Ѿ ´ Ð Ö µ · ¾´ · µÐ¾   Ð for Ð «, The difference between the static and dynamic sub- ¾«¬ ´ ¾ Ó× ½ ´« еµ ¾Ð¬     models is as follows: The target objects can move and ·¾¬ о «¾ «¾ Ô every composite sensor node sends a report at each     for « Ð ¬, ¾ ´Ô ¾ Ó× ½ ´ Ð µ Ó× ½ ´ Ð µµ     measurement epoch. There are no other differences. More precisely, the dynamic submodel is as follows. ·¾ о ¾ · ¾ о ¾ Ô     Each of the ÒÌ target objects may move along an   ¾   ¾   о for ¬ ԾРunknown route with unknown (maybe time-variant) Ô ¾· ¾, speed. Every composite sensor node sends a report at ¼ for · ¾ each measurement epoch. The -th sensor of the -th Ð, composite sensor node sends the report Á ´Ø µ at time (13) Ø , where ½ ¾, ½ ¡ ¡ ¡  . Redefine Á¾ as the Remark 3: When there are multiple rectangular target time average of the number of type- composite sensor objects with side lengths (½ ÒÌ ) satisfy nodes in each of which two sensors detect at least Ð   ¾Ö Ñ Ò´ µ, Eqs. (7), (13), and (9) show È È È½ È Æ ´¾ µ È ´ ´ · ¾Ö µ´ · ¾Ö µ · о   one target object at a single measurement epoch, that ØÑ Ò Ò  ½ ·½ ´Á½ ´Øµ Á¾ ´Øµ ½µ Ñ. ´ È È´ · is Á¾ that ¾Ð ´ · · Ö µµ · ¾´ Ö   Ð µ Ø Ø½ Similarly, redefine Á½ ØÑ Ø Ø½ Ò Ò  ½ ·½ ½ ´ Á½ ´Øµ´½   µ · ´ Ö¾ · о Ð Ö µÒÌ µ . Thus, if we can use   Á¾ ´Øµµ · ´½   Á½ ´ØµµÁ¾ ´Øµ ½µ Ñ, which denotes the Æ ´¾ È withÈ µ various Ð Ö simultaneously, we can es- time average of the number of type- composite sensor timate ´ · µ. However, we cannot estimate nodes in each of which one of two sensors detects at each . Therefore, it is often the case that a pair least one target object at a single measurement epoch.
  • 8. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. IEEE TRANSACTIONS ON MOBILE COMPUTING 8 5.2 Analyzed results for dynamic submodel where ÈÈ Ò Ò Ñ ½ ´Ê Û½ µ ÈÈ is an estimator of Ò , ÒÌ  Ò Ù¾ Û½ µµ ÈÈ ÈÈ ´ ½´ · ѽ Ö ´ ÒÌ  Ò It should be noted that the analyzed results originally ½ ½ Ñ ½ ´Ê Û µ · Û µµ Ò ¡¡¡ ´ ѽ Ö ´ ÒÌ  Ò derived for the static submodel are valid for the dy-  ½ ½ ½ Ѿ ´Ê Û½ µ Û½ µµ Ò namic submodel. The reasons are as follows. (1) At each ½´ · Ѿ Ö ´ ÒÌ  Ò ½ Ѿ ´Ê Û µ · Û µµµ. Ò measurement epoch, the dynamic submodel is identical ¡¡¡  ´ ½ ½ Ѿ Ö ´ to the static submodel. (2) Only the quantity affected by multiple measurement epochs is Á , but it is no 7 N UMERICAL EXAMPLES included in derived formulas. Æ´ µ Á is This section provides numerical examples. The following valid both for the static and dynamic submodels. The conditions were used as a basic pattern for the simula- fact Ú´¢ ϵ ´Á¾ ½ ¡ ¡ ¡ Á¾  µ under the assumption of tion. We used a monitored rectangular area, ¾¼ ¼¼¼ ¢ ½¼¼ no approximation errors or measurement errors is also square units, in which composite sensor nodes were valid. deployed. Three target objects moved at a speed equal to 10 units of length per unit time along a straight line that 6 E STIMATION METHOD was parallel to the bottom line of the monitored area. Based on the analysis in the previous section, we propose Two of the objects were disk-shaped with radiuses of 3 an estimation method for multiple target objects that and 30, and the other one was rectangular with sides may have different parameters. (3, 10). We used six composite sensor nodes of which Note that Á Æ ´ µ for ½ ¾. Thus, Á¾ parameters ´Ð Ö µ were (3, 1), (4, 1), (9, 2), (12, 3), (20, can be an estimator of Æ ´¾ µ . 2), and (22, 1) for ½ . We set ¼ per square unit length for all , and composite senor nodes were Á½ · ½ Æ ´½ µ (16) placed in a homogeneous Poisson process. (As a result, Á¾ · ¾ Æ ´¾ µ (17) the mean density of the sensors was 1 per square unit length.) The mean distance between the target objects where Á   Á is an error of Á from its was 1,000. One simulation yields 2000 measurement expectation. By using Eqs. (6), (7), (8), and (9), epochs, and 10 simulation were run to obtain each result. ÒÌ Á½ · ½ ѽ ´Ì Ð Ö µ (18) 7.1 Approximation errors and sensitivities to vari- ½ ÒÌ ous conditions Á¾ · ¾ Ѿ ´Ì Ð Ö µ (19) We first confirmed the agreement of the simulation re- ½ sults and the theoretically-derived results and evaluated The right-hand sides of these two equations are given approximation errors under various conditions and sen- by derived measures for each class of target object. For sitivities of Á½ (Á¾ ) to various conditions. (In 7.2.1 and example, if the target objects are disk-shaped (rectan- 7.2.2, the impact of these conditions and errors on the gles), Eqs. (10) and (11) (Eqs. (12) and (13)) can be used. estimation accuracy is shown.) We compared Á½ (Á¾ ) ÈÈ ÈÈ When there may be both disk-shaped and rectangular with Æ ´½ µ ( Æ ´¾ µ ), that is, the right-hand side objects, the right-hand sides of Eqs. (18) and (19) should of Eq. (8) (Eq. (9)). For the disk-shaped target objects, Ò ÒÖ be ´ ½ ѽ ´Ê Ð Ö µ · ½ ѽ Ö ´ Ð Ö µµ , Eqs. (10) and (11) were used, and for the rectangular Ò ÒÖ ´ ½ Ѿ ´Ê Ð Ö µ · ½ Ѿ Ö ´ Ð Ö µµ . target object, Eqs. (12) and (13) were used. In general, 7.1.1 Basic pattern Á· Ù (20) For the basic pattern, Figure 2 shows the relative errors Á ´Á½ ½ Á½ ¾ Á½  Á¾ ½ Á¾ ¾ Á¾  µ, of the theoretical values (that is, the relative error = È È where ¡¡¡ ¡¡¡ theoretical value/simulation result -1). Æ ´½ µ shows ´ ½½ ½¾ ¡¡¡ ½  ¾½ ¾¾ ¡¡¡ ¾  µ , and Ù a positive bias because we approximated by for ´ ½ ÒÌ ½ ѽ ´¢ Û½ µ ¡¡¡  ÒÌ ½ ѽ ´¢ Û µ Ú µ. the rectangular target object (see Fig. 1). Æ ´¾ µ also A set of that minimizes the square error Ì ´Á   ¢ can have a positive bias, but it was within a range Ù µ´Á   Ù µÌ can be an estimator ¢ of ¢, where Ì is of simulation error (see Figure 4 for the variance of a transpose operator. the simulation results). In total, the relative errors were ¢ Ö Ñ Ò¢ ´ Á Ù µ´ Á Ù µ Ì (21) small, and we concluded that the theoretical results are valid. When the target object shape can be categorized into sev- eral categories, such as disks and rectangles, the number 7.1.2 Independence to speed, monitored area, and of target objects in each category is also a parameter to moving directions be estimated. For example, when there may be both disk- shaped and rectangular objects, Fig. 3 provides Á¾ when one condition such as the target object speed is modified among conditions used in the ´¢ Ò µ Ö Ñ Ò¢ Ò ´ Á   Ù¾ µ´ Á   Ù¾ µ Ì (22) basic pattern.