ppt for the ieee paper proposed by Jing-Ming Guo and co..

1. Technical seminar on
IMPROVED HAND TRACKING
SYSTEM
Presented by
KISHOR M N
Dept. of CSE
2013/3/6 1

2. Outline
 Introduction
 Existing methods
 Proposed Hand Detection System
 Ada Boosting for real time Hand
detection
 Foreground detection
 Hand Tracking Methodology
 Experimental Results
2013/3/6 2

3. Introduction
 Hand Postures are powerful means
for communication among humans on
communicating.
 Many applications are designed by
using the motion of hand.
 The hand tracking is rather difficult
because most of the backgrounds
change across frames.
2013/3/6 3

4. Existing methods
1) Kolsch and turk proposed a method
to
detect a hand using image cues and
probability distribution.
Method used: Pyramid based
kanade lucas tomasi feature
tracking.
Disadv: It cannot classify wooden
object that are present in the
2013/3/6 4

5. Existing methods
2) Zhu,yang proposed a method in
which each pixel in image is classified
as either hand pixel or background
pixel.
Method used:Hand segmentation
based on Bayes decision theory and
Gaussian mixture model.
Disadv: It cannot distinguish
multitargets in an image and error rate
2013/3/6 5

6. Existing methods
3) Binh and Ejima applied skin color tracking
for face detection to extract the hand
region by separating hand from face.
Method used: Skin color tracking and other
models.
Disadv: It cannot distinguish between skin
region or some object with similar color or
shape as the hand
2013/3/6 6

7. Proposed system
 Local binary pattern (LBP) is one of
the powerful features with low
computation.
 Chen et al.’s work called Haar-like
feature was adopted for hand
detection.
 This paper proposed to combine the
novel pixel-based hierarchical-feature
for AdaBoosting (PBHFA), skin color
detection, and codebook (CB)
foreground detection model to locate a
hand in real time.
2013/3/6 7

8. Proposed system
2013/3/6 8

9. Proposed system
2013/3/6 9

10. Proposed PBH Features
 AdaBoost is employed to select those
few best features from a huge number
of features.
 The PBH features can significantly
reduce the training time for hand
detection than normal features.
2013/3/6 10

11. Proposed PBH Features
Step 1) Given a training positive and negative
samples Xi,Xj of size M × N.
Step 2) Process the samples with the
histogram equalization to reduce the
influence of lighting effect.
Step 3) Calculate the average value Ti of
each sample Xi and use Ti as a threshold
for binarizing the corresponding Xi to
yield a binary image Bi.
Step 4) Given Bi, calculate the probability of
black pixel occurrence at each
position (x, y) and obtain P(x, y).
2013/3/6 11

12. Proposed PBH Features
Step 5) All possible features Fj in a M×N
subwindow can be produced
according to the order table O(x,
y).
Step 6) Finally, these features and the
training image Xi and Xj are fed
into the AdaBoost algorithm.
Each PBH feature can be
considered as a weak classifier.
2013/3/6 12

13. Proposed PBH Features
2013/3/6 13

14. AdaBoosting for Real-Time
Hand Detection
 Aim of boosting is to improve the
classification performance of any
given learning algorithm.
 This phase includes many
classifiers;each weak classifier ht is
made from a feature
ft and threshold Øt
where Pt denotes the polarity used for indicating the direction
of the inequality.
2013/3/6 14

15. AdaBoosting for Real-Time
Hand Detection
• A strong classifier can be obtained by
combining the selected weak
classfiers
from the AdaBoosting and which can
be
recognized as follows
2013/3/6 15

16. Strong classifier
2013/3/6 16

17. HSV Color Space
 A detected hand by the PBH-
AdaBoost algorithm is further filtered
by skin color to reduce false positive.
 HSV color model which is robust to
lighting change is adopted for skin
color localization.
 Advantages of this color model in skin
color segmentation is that it allows
users to intuitively specify the
boundary of the hue and saturation.
2013/3/6 17

18. HSV Color Space
•The hue and saturation are set in
between
0° and 5° and 0.23 to 0.68, respectively,
as
specified in.
2013/3/6 18

19. Foreground Detection
 One of the popular method in foreground
detection is the mixture of Gaussian(MoG).
yet this method has some disadvantages
they are overcome by CB model.
 Kim et al. [2] proposed the CB model for
foreground detection.
 The concept of the CB is to train
background pixel pixelwise over a period
of time. Sample values at each pixel are
clustered as a set of codewords. The
combination of multiple codewords can
2013/3/6 19

20. Foreground
Detection
2013/3/6 20

21. Foreground Detection
 To solve the “still object” problem, a
“buffer” is employed to store the
history of each tracking target. This
buffer is updated frame by frame.
 If one target is detected and tracking,
the value in buffer associates to the
frame that is set as 1.
2013/3/6 21

22. Foreground Detection
2013/3/6 22

23. Hand Tracking Methodology
 Hand tracking phase is the second
step after hand detection.
 Referenced algorithms can be
classified into below categories
1) point tracking
2) kernal tracking
3) silhouette tracking
 Proposed method uses Euclidean
disatnce to track hand.
2013/3/6 23

24. Hand Tracking Methodology
 Eucledian distance is calculated as
follows
 dist1(x, y) < r1.
2013/3/6 24

25. Experimental Results
 In this paper, the public Sebastien
Marcel’s hand posture database [3].
 including “A,” “B,” “C,” “Point,” “Five,”
and “V”
2013/3/6 25

26. Experimental Results
2013/3/6 26

27. Experimental Results
2013/3/6 27

28. Experimental Results
Hand tracking results
2013/3/6 28

29. Experimental Results
2013/3/6 29

30. Experimental Results
Pyramid
based KST
feature
tracking
Appearenc
e based
approach
Proposed
system
2013/3/6 30

31. Conclusion
 Hand tracking system was proposed by
using the PBHFA, skin color segmentation,
and CB model.
 The goal is to use PBH feature to reduce the
required training time and further reduce the
required computation in tracking phase.
 According to the experimental results, the
above tasks were achieved, meanwhile the
tracking accuracy was still maintained in
high level as that of the Haar like feature.
2013/3/6 31

32. References
1) M. Kolsch and M. Turk, “Fast 2D hand tracking
with flocks of features and multi-cue integration,”
in Proc. CVPRW, vol. 10. Jun. 2004, p. 158.
2) K. Kim, T. H. Chalidabhongse, D. Harwood, and L.
Davis, “Real-time foreground-background
segmentation using codebook model,” Real- Time
Imag., vol. 11, no. 3, pp. 172–185, Jun. 2005.
3) Hand Posture Database [Online]. Available:
http://www.idiap.ch/ resources/gestures
4) Jing-Ming Guo, “Effective Hand Posture
Recognition System with Hierarchical-Feature
Adaboosting and Feature Reserving Average
Mask”
2013/3/6 32

33. THANK YOU
2013/3/6 33