Fast lane detection based on bird’s eye view and improved random sample consensus algorithm

Abstract

In order to ensure the safety of drivers, Advanced Driving Assistance System (ADAS) has drawn more and more attention. The Lane Departure Warning system is one of the most important parts of ADAS. However, fast and stable lane marking detection is the precondition of it under complex background. In this paper, we proposed a new lane detection method through bird's eye view and improved RANSAC (Random Sample Consensus) algorithm based on the inspiration that extraction of road features from remote sensed images. According to the bird's eye view of the road image, we can recognize the line marking through Progressive Probabilistic Hough transform instead of lane detection. Then, the detected lines are grouped by a new distance-based weighting scheme and we can get the fields of candidate lanes. For each of the fields, lanes are refined through improved RANSAC algorithm and fitted by double models. Hence, the road orientation can be predicted by the curvature and straight line's slope. At last, our experimental results indicated that the lane detection algorithm has good robustness and real-time under various road environment.