Detection of Road Vanishing Point by Stereo Camera

This paper proposes a novel road vanishing point detection method from a stereo pair of images. First, the improved v-disparity map is generated by sampling the flat region in the u-disparity map. Then the horizon and the approximate road region are extracted by a robust weighted-sampled RANSAC-like method in the improved v-disparity map. With the horizon and road-region constraints, the candidate region of vanishing point can be determined to reduce the complexity of voting. The vanishing point is located at the position with the largest voting score in the candidate region. The proposed approach has been tested on 2621 grayscale image pairs of different road scenes from the KITTI dataset. Our experiments demonstrate that this approach detects horizons and vanishing points at the state-of-the-art performance.