结合Mean-Shift和Particle Filter的鲁棒跟踪算法

Mean-Shift算法在图像跟踪领域得到广泛应用．但有遮挡情况发生时．算法容易陷入局部最大值．Particle Filter作为一种基于贝叶斯估计的算法．在处理非线性运动目标跟踪问题上具有特殊的优势，但该算法计算量大，实时处理能力差。鉴于此，将两种算法相结合，提出一种以重要性函数为切入点将Mean-Shift和Particle Filter相结合的跟踪算法．首先利用Mean-Shift算法跟踪目标，利用目标与模板的相似性系数实时判断，当有遮挡发生时，算法转向Particle Filter进行后续跟踪。实验结果表明，该算法实时性强．跟踪效率高，具有很强的实用性．

Robust Tracking Algorithm United by Combining Mean-Shift and Particle Filter

Mean-Shift algorithm is widely used in image tracking field, but when occlusion occurs, it is easy to fall into local maxium. As an algorithm based upon Bayesian estimation, particle filtering is perdominant on tracking nonlinear moving object, but because of its huge computation, its real-time processing capacity is low. So proposes a kind of tracking algorithm which combines Mean-Shift and Particle Filter by essentiality function. Mean-Shift is used to track object firstly, and real-time judgement is made by the comparability coefficient. If occlusion occurs, algorithm turns to particle filtering to track. Experiment result indicates that the algorithm takes on high efficiency, so it is of high practicability.