Day2-1030-AI-assisted road tracing for OpenStreetMap

Saturday 10:30am, Main hall
Drishtie Patel, Ming Gao

http://2017.stateofthemap.org/2017/ai-assisted-road-tracing-for-openstreetmap/

Transcript

1. OSM @ Facebook Using AI & Machine Learning to Improve OpenStreetMap
2. Product Instagram Places Place Pages Messenger Location Check-In Stories
4. Road Coverage %
5. Process
7. Challenges
8. Training Data
9. ML Output Mask: Raw Detection
10. Post Processing: Threshold + Centerline
11. Post Processing: Trim, Connect, Merge
12. Post Processing: Remove Island Roads
13. Convert into OSM Format
14. Roles Editor Validator Submitter
15. Human Validation
16. Community
17. Road Density Import Progress
18. Road Density 13 of 79 Provinces 2,263 of 17,574 Tasks 76,434 New Ways 37,685 km’s of road 1,044,820 Nodes Stats from May 2017 Road Density
19. Thailand Before/After Before After
20. Road Type • Inference from satellite imagery and topology • Studying local OSM edits • OSM Wiki’s • Talking to the community • Population density maps
21. Population Density Data used for Tagging
22. Crowdsourcing Road Names Red Bird - Kamphaeng Phet 2 Road Kamphaeng Phet 2 Road Matching FB place addresses to road Name candidate extraction
23. Quality Analysis • Customized iD Editor • JOSM Validation • OSMCHA • OSMOSE • KeepRight
24. Successes • Technology • Improved road geometry quality • Fine tuned overall processes • Community engagement • New Imagery in OSM Challenges • OSM process for imports • Lag time between editing and submission • No local community?
28. Example 5
33. Drishtie Patel osm@fb.com Would love to hear your thoughts on ways to collaborate and support the community ☺