Multi-sensor Based Lane Detection and Object Tracking Method

Considering the relation between the movement of vehicles and the lane parameter, a novel fusion method of lane detection and object tracking is proposed. Though millimeter radar can accurately provide longitude range and velocity information of vehicle ahead, it can not recognize lateral position and road state, which makes it easy to loss targets when vehicle ahead turns or changes its lane. To solve this problem, lane information achieved from image is integrated with radar-filtered information. With the selected road shape model and the intensity feature of lane image, an optimization algorithm was established to maximize likelihood function evaluating how well the image gradient data on an assumed lane marking supports a given set of template parameters. Lane curve parameter can be transformed into vehicle front wheel orientation, and finally the fusion estimation of vehicle state can be achieved by UKF (uscented Kalman filter) estimator and least-squares based data fusion algorithm. Simulation results validate the proposed method can improve vehicle´s pose tracking accuracy significantly.