A vision system for autonomous vehicle navigation in challenging traffic scenes using integrated cues

This paper presents a system used for navigating an autonomous vehicle through challenging traffic scenes based on stereo vision. In order to safely as well as legally facilitate autonomous driving behaviors such as road following, obstacle avoidance, off-road navigation, etc., the proposed system consists of two parallel modules: drivable road region detection and tracking module and painted lane boundary detection and tracking module. In these two modules, both intensity and geometry cues of the road scenes are utilized and integrated for detection and tracking of the targets in probabilistic models to deal with challenging conditions and situations. The obtained results are subsequently correlated for navigation. Furthermore, the geometry relationships between the stereo camera in the moving vehicle and the road are dynamically estimated and calibrated, and the parameters in the probabilistic models are unsupervisedly learned from the input image pair itself so as to adapt to changing environments. Therefore, more accuracy and robustness can be expected in the proposed system. Experimental results in various real challenging traffic scenes show the effectiveness of the proposed system.