Mean-shift based visual tracking algorithms have several desirable properties, such as computational efficiency,few tuning parameters, relatively high robustness in performance and straightforward implementation, which make them to become an appealing topic in visual tracking research area. Firstly, original mean shift tracking algorithm was introduced and its defects were pointed out afterwards. hhen improvements of the original algorithm were elaborately discussed from five aspects, namely generative and discriminative object appearance model, model update mechanism,scale and orientation adaptation, anti-occlusion and fast moving object tracking. Both classical algorithms and recent advances are included in each aspect. Finally,the prospects of mean-shift based tracking were presented.