Vehicle Occlusion Detection Algorithm

Object tracking is a very important branch of compter vision, the essence of which is automatic recognizing moving target, deciding the position of the target and automatic tracking the target. Object occlusion often appears in tracing and recognizing the moving target based on image. This is a difficulty to solve the problem in dynamic image. Especially vehicles occluding each other within the image plane have caused problems with the measurement accuracy. Hence it is necessary to establish a reliable, practical mechanism for the elimination of the occlusion. In this paper, moving object tracking is investigated and special attention is paid to occlusion handling.First, we need to extract the very satisfying vehicle prospect for occlusion handling. For vehicles in the detection of high-definition images, the existing background modeling methods extract the vehicle prospect with too much computation, while exists empty and chasm in the detected vehicle regions. This article proposes a prospect-extraction algorithm based on the high-definition images. First of all, we need to remove the interference region to reduce the computation by setting the region of interest. Secondly, we will definite vehicle region through the background difference algorithm and edge detection, then contour tracing of the contours to extract the complete contour of the vehicle. The experimental results show that the algorithm can extract the vehicle prospect completely with higher speed compared with existing methods such as GMM and Bayes, and can apply to high-definition graphics applications.Then, This article proposed two algorithms to handle the partial-occlusion between two vehicles.One of the algorithms is a contour-based vehicle partial-occlusion detection algorithm. Once the moving vehicles are detected,Then we carry on the contour and convex hull analysis to the detected vehicles. Then vehicle occlusion is detected by using the region area relations of the vehicle's contour and its convex hull. Once occlusion is detected, an optimal cutting line algorithm is implemented to divide the occluded vehicles into two vehicles. This approach needs no prior models other than the shape of vehicles. The experiment results have shown that the proposed algorithm is effective to deal with the partial vehicle occlusion.Another method is based on the positioning of the windows of the occluded vehicles. When two occluded vehicles were detected as the same car by mistake in the object tracking, generally speaking, their windows are not occluded. First, we can draw a brightness curve on the gray image, then extract the window's location by analyzing the light curve. Once the windows are positioned, we make a dividing line in the appropriate position, thus the occluded vehicles are separated and the accuracy of object tracking is enhanced. The experiment results have shown that the proposed algorithm is is simple and effective, a large number of false information of detection of vehicles can be amended.