Introduction
Goals
The extraction of a street grid in dense urban
environments producing a description of
extended streets and their intersections.
- Model based extraction, data fusion, and multiple views.
- Initial effort on regular patterns using models of intersections.
- This complements much of the other work in road detection.
Where does this fit in the spectrum of road detection algorithms:
- Low Resolution Rural roads.
- Low Resolution Rural roads. Best exemplified by the SRI
road work.
- Few intersections.
- Road extracted separately, connections are not important.
- High resolution Road Following
- Folow the intensity profile of the road.
- Automatically find seeds and extend from the seed.
- TUM, Raffael/Brown
- Refinement of user inputs
- Snakes (SRI, Ziplock snakes, etc.)
- Regular urban street grid patterns. Stereotypical Midwestern US.
- Combinations of regular grids. Change in the grid, but locally regular.
- Irregular street patterns. (curves, circles, etc.)
- General urban streets (e.g. Boston, other older cities)
- Occlusions from adjacent buildings (dense urban settings -- New York)
Generally the research efforts do not generate a topological description
of the road network, they generate spaghetti-like roads.
Assumptions
- Simple three-dimensional road segment and intersection model.
- Known camera parameters
- Either nadir or oblique views are possible
- Roads have visible edges
- Regular Street Grid
- Detect when regular grid ends
Maintained by
Keith Price,
price@usc.edu