Vehicle lateral position estimation method based on matching of top-view images

In this paper, a method to estimate the lateral position of the vehicle from a sequence of moving camera images is proposed. Proposed method relies on the plane projective transform (Homography) between the ground and the image plane of the present and the next frame. Homographies are obtained according to the information of the calibrated camera and the vehicles' speed. The movement of the camera is obtained by the registration of the 2 consecutive input images. The proposed method does not rely on extraction of features such as lines, flow vectors or lane markers, but based on matching of warped top-view images between two consecutive frames. Therefore, vehicle movement can be estimated even from images without such explicit image features or without special sensors or GPS. In this paper, experiments are done with both synthesized and real images and enough accuracy is shown even compared to conventional method.