在对Mean Shift跟踪算法原理分析的基础上提出了一种反映区域像素空间关系的特征值与原算法基核结合的跟踪方式。该方式采用适应性公式与相似性量度结合的二级判决策略，提取区域内特征值并按不同权重归一化特征值。若不满足特征观测值的条件，则采取粗-精搜索结合的方式对目标模型进行更新。实验结果表明目标跟踪过程中提取的特征值波动符合目标区域实际情况，并在波动异常时重新定位目标，从而提高了跟踪的稳定性。

A tracking mode is presented which reflects regional pixel spatial relationship combining with the core of the original algorithm based on the analysis of the principle of algorithm. Meanwhile the adaptability and the similarity measure formula with two level decision strategy are adopted，and it extracted region characteristic to normalize values by the different right. If it does not meet the characteristics of observed value condition, the coarse-fine search to update the target model should be adopted. Experimental results show that the fluctuations characteristic observations meet the actual situation of the target area in the target tracking process, and the target in the case of abnormal fluctuations can be repositioned, thereby the tracking stability can be improved.