Dynamic Determination of Histogram Resolution for Object Tracking using the Mean-Shift Tracking Algorithm

Takayuki Nishimori, Toyohiro Hayashi, Shuichi Enokida, and Toshiaki Ejima


Tracking, Image Understanding, Meanshift, Histogram


An object tracking is one of the most important technique in digital image processing. The mean-shift tracking algorithm can be regarded as a specialization of template matching that scans locally according to the gradient of color features so that targets can be estimated faster. However, the mean-shift algorithm only guarantees a locally optimum solution, not a globally optimum solution. Thus, it is difficult to re-track the target automatically once tracking has failed. In this paper, to suppress the local optimal solution, a controlling method of the color histogram representation is proposed. The proposal method controls the color representation not to make the color histogram of a target area similar to background areas histograms. Actually, some experimental results showed the effectiveness of the proposal method.

Important Links:

Go Back