An improved CamShift algorithm using stereo vision for object tracking

Stereo matching is a powerful method of image segmentation for applications such as moving objects tracking. The result of a stereo matching algorithm is a disparity image which shows the difference of the object location in the images produced by the left and right camera. In this paper, we present a tracking algorithm which is based on the CAMSHIFT (Continuously Adaptive Mean Shift) algorithm. Our algorithm operates on the stereo images, and the disparity image. Extending the CAMSHIFT algorithm with the scene depth information from the disparity image we are able to increase the tracking quality with acceptable losses in the execution time. The algorithm is implemented and evaluated in a people tracking system, where the experimental results show that our algorithm outperforms the conventional CAMSHIFT algorithm in tracking accuracy.