中文摘要：

针对传统Meanshift 算法对快速运动目标跟踪不准和持久跟踪易丢失问题，提出一种新的Meanshift快速运动目标鲁棒跟踪方法。新方法结合Kalman滤波预测从帧差法检测的运动区域中挑选可信区域只进行一次Meanshift 颜色相似检测，若相似值不满足条件则再从其他运动区域依次进行原Meanshift跟踪，找出最佳跟踪区域。新方法减少了原方法颜色匹配迭代次数，对目标持久跟踪丢失时也可以快速重新找回原目标进行跟踪。最后，跟踪实验结果也说明了新方法计算更快和跟踪上的鲁棒性。

英文摘要：

Traditional Meanshift algorithm has problems in inaccurate tracking of fast-moving targets and easy to lose the targets enduringly tracked. In light of these, in the paper we present a new Meanshift robust tracking method for fast moving target. The new method combines Kalman filtering prediction and selects trusted regions from moving regions detected with frame difference method and carries out Meanshift colour similarity detection only once, if the similarity value does not meet the condition, then it conducts original Meanshift tracking in turn from other moving regions and finds out the best tracking region. The new method decrease the iteration times of colour matching in previous method, and can fast find back the original targets to track again once losing them during a long period tracking. In end of the paper, the results of tracking experiments also show that the new method computes faster with robustness in tracking.