基于SVM数据融合的实时粒子滤波算法

采用粒子滤波的目标跟踪算法在粒子数目较多时计算量大、实时性差，针对该问题提出了一种新的基于支持向量机数据融合的实时粒子滤波算法。该算法在估计窗实时粒子滤波的基础上，使用支持向量机融合窗内不同时刻粒子集，并根据融合的结果更新粒子权值，实现对目标状态的快速跟踪。相对于原算法采用最小化Kullback-Leibler距离来调整估计窗混合分布的权值，该方法的计算复杂度低、速度快，进一步提高了算法的实时性。对纯角度目标跟踪问题的仿真结果表明了该算法的可行性和有效性。

Real-time particle filter based on data fusion with support vector machines

To overcome the drawback of high computational burden and poor real-time capability in target tracking problems using particle filter with a large number of particles, an improved real-time particle filter (RTPF) algorithm is proposed which is based on data fusion with support vector machines (SVM). The SVM RTPF employs the estimation window RTPF as basic framework and uses the SVM for fusing the particles at different time in the window, so the target is tracked quickly by the particles and their updated importance weights according to the fused results. Compared with the RTPF algorithm based on minimizing the Kullback-Leibler distance to adjust mixing weights in the window, the new algorithm is simple and more suitable to the range of real-time applications. The bearings-only tracking simulation results demonstrate the feasibility and superiority of the novel algorithm.