Road recognition in urban environment

摘要

The position estimation of a vehicle relative to the lane and the recognition of the road topography are important tasks for vision-based driver assistance in urban traffic. Lane structures such as markings, curbs, crosswalks and stop lines are detected in a polygonal edge image. Data driven global detection and fast, local model-based tracking are integrated in the presented lane recognition system that reduces the computational effort and keeps up with the dynamic urban environment. Stereo vision is integrated to improve robustness in cluttered scenarios and to reliably estimate horizontal and vertical geometry and motion.