A shadow elimination method for vehicle analysis based on random walk

A novel method is proposed for solving the shadow and occlusion problems of vehicle analysis. Kalman filter is combined with random walk algorithm. First, the computation region of random walk is reduced through the prediction information from the Kalman filter, then the seed points is extracted in this region for segmentation. Further, the segmentation of random walk is implemented, and the results of which is used to update the filter parameters. In order to obtain the initial state vector for Kalman filter, the random walk based on car bottom shadow is proposed too. Experiment results show that the problem of moving vehicles shadows, tracking and occlusion can be solved.