1. Improving Navigation Automated Name Extraction for
Separately Mapped Pedestrian and Cycle Links Anita Graser, Markus
Straub This work is partially funded by the Austrian BMVIT
programme Mobilitt der Zukunft under grant 844434 PERRON as well as
the Vienna Business Agency call From Science To Products 2013 under
the grant for project sproute.
2. Outline 1. Problem description 2. Approaches 3. Results 4.
Discussion 5. Outlook 2
3. Unnamed Link Naming Problem (ULNP) finding which named
street an unnamed pedestrian or cycle link belongs to 3 Follow the
cycle path along Burgring
4. Motivation 4
5. Motivation 5
6. Motivation 6
7. Motivation 7
8. Motivation 8
9. Automatic Name Extraction 9
10. Methods Hausdorff distance matching Median distance
matching Composite matching: distance and orientation 10
11. Hausdorff distance matching Best match = smallest Hausdorff
distance + check distance tolerance 11 (, ) = , = { , , , } , = max
{ min { , }}
12. Median distance matching Best match = smallest median
distance + check distance tolerance 12 (, ) = median { min { , }}
.
13. Composite matching Best match = best combination of
distance and similar orientation + check distance and angular
tolerance 13 , = median { min { , }} + (, ) 180
16. Hausdorff distance matching = correct name no match wrong
name sum already named 26 (100%) 26 should be matched 252 (66.7%)
67 (17.7%) 59 (15.6%) 378 should not be matched 338 (84.5%) 62
(15.5%) 400 17
17. Hausdorff & Median matching 18 Hausdorff distance
matching Median distance matching
18. Median distance matching 19 = correct name no match wrong
name sum already named 26 (100%) 26 should be matched 284 (75.1%) 2
(0.5%) 92 (24.3%) 378 should not be matched 308 (77.0%) 92 (23.0%)
400
19. Median & Composite matching 20 Median distance matching
Composite matching
20. Composite matching 21 = , = correct name no match wrong
name sum already named 26 (100%) 26 should be matched 350 (92.6%)
20 (5.3%) 8 (2.1%) 378 should not be matched 353 (88.3%) 47 (11.8%)
400
21. Results 2208.07.2015 50% 55% 60% 65% 70% 75% 80% 85% 90%
95% 100% Hausdorff distance Median distance Composite Percentage of
correct matches should be matched should not be matched
22. Composite matching issues 23 Extensions
23. Composite matching issues 24 1:n matches
24. Conclusion Solving the ULNP Composite matching succeeded in
matching 90.7% of test links better performance expected for
rectangular street networks at roundabouts, a local relaxation of
the angular tolerance might lead to better results. Potential 2nd
application: street graph generalization enrich the generalized
link with attributes from all matching links automatic inference of
street cross-section characteristics
25. Future work address shortcomings of Composite matching
identify situations where links have to be split to be able to
compute appropriate matches Test transferability to other
cities