Automatic Road Finding

Algorithms have been developed for completely automatic detection and accurate estimation of road boundaries in aerial visible light imagery. A low-level algorithm uses dynamic programming to realize maximum a posteriori probability estimation of the boundaries within small windows to provide some initial estimates, and then a multi-hypothesis generalized Kalman filter runs in real-time to estimate entire roads in the presence of modest amounts of clutter, missing edges, vehicles on the roads, road intersections, occlusion due to over-passes, etc. Work is continuing on increased capability. Software is available.

Back to research topics.