Inverse perspective mapping for real-time Lane Detection in City Streets

Lane detection is a crucial step in any autonomous driving system in order to decrease car accidents and increase safety. In this paper, based on inverse perspective mapping and Probabilistic Hough Transform, we propose a lane detection system which works on city street images. First, we calculate the top view of street image by adopting inverse perspective mapping. Second, the lanes are rectified using a specifically designed filter which enhances the lanes and suppresses other elements. Then, by using Probabilistic Hough transform the location of the lanes is detected in the images. For the final refinement, lane candidates are mapped to the road image using perspective mapping and the lane intensity is analyzed to reduce false acceptance. We evaluate the performance of the proposed method on Caltech-lane dataset and the obtained results prove that the proposed algorithm is able to detect straight lanes.