Abstract:

The paper proposes a new method of combining perspective points and vehicle contours to segment the occluded vehicles in the image for resolving the problem of multivehicle occlusion in vehicle detecting and tracking. Perspective points are detected by analyzing the lane lines in the highway scene, and the cuboids of vehicle models are built with the perspective points. The ratio of the cuboid’s height to width and the size of the vacant area inside the cuboid models are used to detect the occluded cars. Different searching methods are chosen to find the cutpoints, depending on the condition of occlusion. Then, the occluded vehicles are detected and segmented in sequence from front to back. The experiments show the accuracy and adaptability of the proposed method in the segmentation of the occluded vehicles on a straight road.

Key words: vehicle occlusion；segmentation；cuboid models；contour