Vision-based lane line detection for autonomous vehicle navigation and guidance

Abstract

Vehicle pose estimation with respect to the road plays a critical role in the advances of autonomous vehicle navigation and guidance. Vision-based road lane line detection provides a feasible and low cost solution as the vehicle pose can be derived from the detection. While good progress has been made, the lane line detection has remained an open one, given challenging road appearances. In this paper, we propose a more robust vision-based approach by making use of ridge detector and sequential RANSAC (RANdom Sample Consensus). The pose estimation accuracy and consistency are improved by imposing parallelism constraints and fitting multi road models simultaneously. The algorithm is so robust that it is still able to work even when lane line only exists on one side of the road.