基于混合粒子滤波的目标跟踪

为了克服多目标跟踪中估计效果对初始样本选择的强依赖性，首先通过Harris角点检测和KLT算法实现对图像序列中的特征提取和匹配，然后利用Mean-shift算法对匹配的特征点进行聚类和定位，将Mean-shift算法与粒子滤波器相结合，提出了基于Mean-shift算法的混合粒子滤波器，给出了具体算法流程，并就实际图像序列的动态多目标跟踪进行了实验，实验结果证明了该方法的有效性。

Targets tracking based on mixture particle filtering

Multiple targets tracking is an important task in the field of computer vision. To overcome the strong dependence of evaluating effects on the initial selection of the sample, Harris corner detector was used to detect the feature point in the frames, at the same time, KLT tracker was used to match the feature point in the consecutive two frames; Then, Mean shift algorithm was applied to cluster and locate the feature points which were matched by KLT algorithm. Furthermore, Mean-shift algorithm was combined with the particle filter, and a mixture particle filter based on Mean-shift was proposed. The detailed algorithm was given in this paper and it was applied in the multiple targets tracking. Finally, the experiment has testified the validity of the method.