Traditional Mean-shift algorithm for object tracking has some disadvantages

such as the localization drift caused by the background pixels and the tracking failure when the object undergoes occlusion. In order to overcome the above mentioned 2 shortcomings

improved Mean-shift algorithm is proposed. Firstly

according to the difference of color distribution between the object and the background in the initial frame

log-likelihood image is set up to select the discriminative color features for object modeling. The candidate modeling is done the same way. Secondly

the whole candidate region is separated into several overlapped fragments. Mean-shift iteration is done to every fragment and the target localization is reset by the location of the fragment in the candidate region that matches most to the corresponding fragment in the object region. The fragment based Mean-shift is very robust to partial occlusion. When object is severely occluded

linear prediction is used to estimate probable location of the object in the next frame. Experimental results prove that tracking using the improved Mean-shift algorithm has good localization precision and is robust to partial and severe occlusion.