中文摘要：

针对多目标跟踪中相似目标的发散问题和跟踪核函数窗宽固定的缺陷,提出一种基于FCM（fuzzy C-means）聚类的粒子滤波算法。该算法结合经典粒子滤波理论,使用可变椭圆作为粒子区域,在粒子滤波的重要性重采样后,通过Mean-Shift算法获得每个目标的聚类中心,使用FCM聚类算法完成粒子聚类,获得相应目标的粒子子群,最后通过粒子子群估计各目标的最终状态并修正核窗口宽度。实验表明,与传统粒子滤波算法相比,该算法解决了传统粒子滤波的发散问题,减少了粒子数量,能够准确地对多目标进行跟踪,具有很好的鲁棒性和实时性。

英文摘要：

Divergence problem of similar targets and fixed window width of target kernel function are the defects existing in multiple-object tracking method.The new algorithm stated in this paper is based on classical particle filter theory.Firstly,variable ellipse is used as particle zone to gain the cluster center of each object through Mean-Shift algorithm after importance re-sampling.Then,FCM is used to complete particle clustering and get the particle subset of respective objects.Lastly,the final state of each target is calculated and the kernel window width is revised through particle subgroup.Experiments prove that this algorithm can solve the divergence problem of traditional particle filter,reduce number of particles,and has robustness and real-time property.