Real-time lane detection for intelligent vehicles based on monocular vision

A novel real-time lane detection system which is capable of detecting multiple and curved lanes rapidly is proposed in this paper. Warp Percpective Mapping is used firstly to generate a bird's view which can get rid of the perspective effect. A fast curvilinear structure extraction method is applied in the lane detection sytem to retrieve an accurate set of lane pixels, providing not only the pixels' positions, but also the directions. This feature distinguishes the method which differs from others. An improved Hough Transform is followed to mark the initial lane shapes and locations. After the line refining and extension steps, a fast and robust RANSAC style algorithm is applied to shape the lanes with third degree Bezier Spline fitting. Experiments tested on the video sequences show that the algorithm can detect all multiple lanes in the scene and achieves a high detection rate with a very low time cost. The results analysis is carried out, emphasizing the comparable performance to previous methods.