Pedestrian navigation systems available today still use paradigms developed for car navigation. We present a novel landmark-based pedestrian navigation model using only OpenStreetMap data, which is open and available globally. This approach ensures that our landmark navigation model is widely applicable, rather than restricted to a certain area with exceptional data sources. Our contributions cover algorithms for extraction, weighing, and selection of landmarks based on their suitability, as well as the generation of landmark-based navigation instructions for pedestrian routes. Initial tests with pedestrians show promising results by confirming that our weighted landmark selection reduces the number of navigation errors and revealing future challenges for the generation of intuitive pedestrian navigation instructions.

1. Anita Graser
Scientist, Center for Mobility Systems – AIT Austrian Institute of Technology
Plaza Mayor – cc-by Kris Arnold on Flicker
LANDMARK-BASED INSTRUCTIONS FOR
PEDESTRIAN NAVIGATION SYSTEMS USING OSM

2. We want to …
 Compute realistic pedestrian routes
 Compute pedestrian navigation instructions with landmarks
 Use only commonly available data, i.e. OSM
OUR POINT OF DEPARTURE

3. 1. Constructing a pedestrian routing graph from OSM
2. Extracting landmarks from OSM
3. Generating landmark-based pedestrian instructions
OUTLINE

4.  using information from globally available OpenStreetMap database
 based on sidewalks not on roads
 considering way quality criteria (based on selected route attributes)
 including road crossing and square crossing
TOPIC 1: CONSTRUCTING A PEDESTRIAN
ROUTING GRAPH FROM OSM
431.05.2017
Naumann S. and Kovalyov M.Y. (accepted) Pedestrian Route Search based on OpenStreetMap. 3th SCIENTIFIC AND TECHNICAL CONFERENCE
TRANSPORT SYSTEMS THEORY AND PRACTICE, Katowice, Poland, September 19-21, 2016.
Naumann, S., Czogalla, O., Kühner, F. (2015) A Safety Index for Road Crossing, Future Active Safety Technology Towards zero traffic accidents, FAST-
zero 2015 Symposium, Gothenburg, Sweden.

5. INTEGRATION OF PEDESTRIAN SQUARES
INTO THE ROUTING GRAPH
531.05.2017Graser, A. (2016) Integrating Open Spaces Into OpenStreetMap Routing Graphs for Realistic Crossing Behavior in Pedestrian Navigation. GI_Forum
‒ Journal for Geographic Information Science, 1-2016, 217-230, doi:10.1553/giscience2016_01_s217.

6. experimental implementation in QGIS
ROUTING ON THE PEDESTRIAN GRAPH
31.05.2017

7. TOPIC 2: EXTRACTING POTENTIAL
LANDMARKS FROM OSM
731.05.2017
OSM import
with ogr2ogr
PERRON views

8. POINT LANDMARKS
31.05.2017
…

9. POLYGON LANDMARKS
31.05.2017
…

10. WEIGHTS FOR LANDMARKS
1031.05.2017
Descriptive
name
Weight
…

11. Pedestrian-centered navigation instructions
 using information from globally available OpenStreetMap database
 automatic selection of most suitable landmark
TOPIC 3: GENERATING LANDMARK-BASED
PEDESTRIAN INSTRUCTIONS
1131.05.2017
Turn left
after Café
Hegelhof

12. 1. Extracting decision point candidates along the route
2. Filtering & merging of decision point candidates
3. Determining crossing types
4. Splitting the route into episodes between decision points
5. Selecting landmarks for decision points
6. Computing instructions per episode
7. Generating internationalized text for instructions
INSTRUCTION CALCULATION PROCESS

13. Possible reasons are:
 Start of the route
 Mode of transport change
 Direction change
 Name change
 Functional Road Class (FRC) change
 Form of Way (FOW) change
 Road crossing (at start of each link that is a crossing)
 Begin / end of stairs, bridges, tunnels, roundabouts…
STEPS 1 – EXTRACTING DECISION POINTS

14. STEPS 2 - REMOVING UNNCESSARY DECISION POINTS

15. STEP 3 – CROSSING TYPES

16. Splitting  episodes between decision points
STEP 4 – SPLITTING AT DECISION POINTS

17. Approaches
 All named buildings in buffer [Elias & Sester 2002]
 Most unique building at decision point [Elias 2003]
 Rule-based using categories [Duckham et al. 2010, Dräger & Koller 2012]
 Features with highest „landmarkness“ [Quesnot & Roche 2014]
 Neuronal network [Zhu & Karimi 2015]
STEP 5 – SELECTING LANDMARKS

18. 𝑆 = ( 𝑑 𝑚𝑎𝑥 − 𝑑 ∗ 𝑤 𝑑 − 𝑐 𝑚𝑎𝑥 − 𝑐 ∗
𝑑 𝑚𝑎𝑥
𝑐 𝑚𝑎𝑥
∗ 𝑤𝑐 + 𝑠 ∗ 𝑤𝑠 + 𝑙 ∗ 𝑤𝑙) ∗ 𝑣
where
 d is the distance between decision point and landmark,
 dmax is the maximum distance for a candidate to be considered,
 c is the landmark category weight,
 cmax is the maximum landmark category weight,
 s is the side of the landmark relative to the next turn: same side (1) or other side (0),
 l is the location of the landmark relative to the route: before (1) or after (0) the decision point,
 v is the visibility of the landmark: visible (1) or hidden (0), and
 wd, wc, ws, wl are the weights for the terms for distance, category suitability, side, and location.
STEP 5 – SELECTING LANDMARKS
Land-
marks
10 nearest point
& 10 nearest polygon LMs
LM candidates
sort by score
Best LM

19. STEP 6 GENERATING INSTRUCTIONS

20. Handling micro mapping …
STEP 6 GENERATING INSTRUCTIONS

21. # 0=landmark, 1=on street
RoadInstruction.ROUTE_START = Start {0} {1}
Preposition.BEFORE = before the {0}
Preposition.AT = at the {0}
Preposition.AFTER = after the {0}
amenity=library = library
RoadInstruction.onto_start_street = on {0}
STEP 7 INTERNATIONALIZATION
# 0=landmark, 1=on street
RoadInstruction.ROUTE_START = Starten Sie {0} {1}
Preposition.BEFORE_m = vor dem {0}
Preposition.BEFORE_f = vor der {0}
Preposition.BEFORE_n = vor dem {0}
...
amenity=library = die Bücherei
RoadInstruction.onto_start_street_m = auf dem {0}
RoadInstruction.onto_start_street_f = auf der {0}
RoadInstruction.onto_start_street_n = auf dem {0}

22. OPEN ROUTE FORMAT
2531.05.2017
https://github.com/dts-ait/sproute-json-route-format

23. http://bit.do/perron
DEMO WEBSITE

24.  Further improvement of navigation instructions
 Field tests in Vienna and Magdeburg
 Calibration of weights for landmark selection
algorithm
ONGOING AND FUTURE WORK

25. CONTACT
Anita Graser – anita.graser@ait.ac.at
@underdarkGIS