Ground plane rectification based on rich line representation of vehicle in surveillance

Outdoor visual surveillance scenes usually contain lots of objects moving on a ground plane. However, the perspective distortion brings in the result that the same object moves faster and looks larger when it is close to the camera, which makes the primary surveillance scenes pictures can't be used for further research directly. For example, accurate map-making, precise measurement of distance or angles, 3D model estimation and recovery and so on. Therefore, some kind of methods should be provided to eliminate the perspective distortion. In this paper, we make full use of target recognition such as moving vehicles in a video to accomplish rectification. First, we separated the moving targets from the background, then, we detected a lot of line segments from the moving vehicles in each frame, and calculated the vanishing points with parallel line segments and calculated the affine matrix with perpendicular lines, and then, we performed linear regression on the vanishing points and get the vanishing line, at last, we have the perspective matrix and affine matrix calculated and do the rectification to the whole surveillance scene.