Robust multilane detection and tracking in urban scenarios based on LIDAR and mono-vision

Lane detection and tracking is the basic component of many intelligent vehicle systems. In this study, a robust multilane detection and tracking method is proposed. Using the measurements provided by an in-vehicle mono-camera and a forward-looking LIDAR, this algorithm can address challenging scenarios in real urban driving situations. The proposed approach makes use of steerable filters for lane feature detection, LIDAR-based image drivable space segmentation for lane marking points validations and the RANdom SAmple Consensus technique for robust lane model fitting. To improve the robustness of the fitting further, the parallel lanes hypothesis is introduced. The detected lanes initialise particle filters for tracking, without knowing the ego-motion information. The image processing procedures are carried out in inverse perspective mapping image, because of its convenience for multilane detection. Experimental results indicate that the algorithm in this study has robustness against various driving situations.