Research On Mean Shift Target Tracking Algorithm Based On Window Extracting Automatically

Mean Shift algorithm need to select its initial tracking window manually for target tracking, when speed of target is too fast or the target scale is too small, the mode of selecting manually often appears the problem of missing the target, the target center deviated from the center and window contains too much background information. Aiming at above problems, an initial tracking window automatic selection method is proposed in this thesis. This method apply gaussian mixture model which improves the update mode of learning rate and the determination method of background model threshold to detect moving target, using the horizontal projection and vertical projection to split candidate target region after removing the noise in the foreground image, improved chain code tracking is executed in each candidate tracking region, the initial tracking area is determined through statistical chain code point number, according to the coordinate of chain code point to calculate the tracking window size and center position coordinate. Apply this window extraction automatically methods in Mean Shift target tracking algorithm and make experiment, the experiment results show that, compared with the method of selecting the tracking window manually, this method can timely, accurately extract the initial tracking window, and window contains less background information, in the condition of frame rate of 25 frames per second, can realize real-time, accurate tracking.