Real-time Robust Multi-lane Detection and Tracking in Challenging Urban Scenarios

In this paper, we consider multi-lane detection in challenging urban scenarios such as emerging, ending, spitting and merging of lane markings, heavily curved lanes, zig-zag lanes, on/off ramp and disturbance of other road writings. We present a fast robust multi-lane detection and tracking framework to address these challenges. In this method, lane feature elements are first extracted and then grouped into clusters, and clusters are associated through energy minimization. Probabilistic decision making is adopted to track individual lane considering lane cluster measurements and prior lane state. A multi-lane tracking strategy is also presented to manage lane tracks from their appearance to disappearance, which can reduce false detection and improve robustness of the algorithm. Real driving data are used to verify the effectiveness of our algorithm in all mentioned challenging scenarios.