A Multi-Video Vehicle Tracking Method Based on Camshift with Color Interference

Abstract

To avoid the situation that the Camshift algorithm which based on color characteristic will lose the target vehicle when other similar-color vehicles appear in the same video, we propose a method that can apply to multi-video tracking. Firstly, set Bhattacharyya distance at 0.8 as the standard of a successful tracking. Then in continuous frame image, use the Kalman filtering to estimate the next position of the target in order to narrow the tracking range and avoid losing target. In the case of discontinuous frame image, we suggest that use the color characteristic of the camshaft to track at first. If tracking fails due to color interference, then use the SIFT to track targets. Based on the above, we can achieve multi-video tracking. The experiment shows the method has the accuracy rate of 94.8% in continuous frame image and 91.4% in discontinuous frame image under a poor environment excluding a poor visibility.