A lane detection algorithm based on wide-baseline stereo vision for advanced driver assistance

In this paper, we present an efficient lane detection and tracking algorithm for wide-baseline stereo vision systems with cameras mounted on the host vehicle. The main purposes of the vision system are host vehicle localization and lane shape recognition for advanced driver assistance tasks on highways and country roads with visible lane boundary markings. This is achieved via the 3D geometrical reconstruction of the detected markings. Although, the wide baseline results in an increased accuracy of lane reconstruction, it also increases the local dissimilarities in corresponding images, thus, rendering the stereo matching problem more difficult. A model-based feature matching method is used to solve the correspondence problem. A fast ground plane estimation is followed by the horizontal lane profile extraction with a robust multi-stage fitting of a parabola-pair lane model on the metrically reconstructed lane boundary features. Reconstructions from real-traffic images are presented.