Machine Vision to Lane Departure Sensing System for Urban Traffic

A lane departure sensing system with uncalibrated camera is proposed in this paper. The study focuses on urban traffic scenes with varieties of lane marks instead of simple highway scenes. A distance map and non-symmetrical 2-D Gaussian distribution matrix are used to generate various voting value for Hough transform, so that the lane marking and lane signs are excluded and the accuracy of lane boundary detection is improved. Lane width and lane boundary continuity are obtained for better reliability and quality of lane detection for urban traffic scenes. The result of lane boundary detection is then used to establish a lane departure sensing system, which can be used to enhance driving safety. Experimental results of urban road scenarios have demonstrated feasibility and robustness of the proposed method.