基于改进概率假设密度的多目标跟踪算法

经典序贯蒙特卡罗概率假设密度(Sequential Mote Carlo Probability Hypothesis Density, SMC-PHD)滤波中, 将目标状态转移密度函数做为建议密度函数, 没有利用当前观测信息, 导致大部分预测粒子状态偏离目标真实状态, 粒子退化严重.针对上述问题, 提出利用均方根容积卡尔曼滤波产生建议密度函数, 对其进行采样得到预测粒子状态, 该方法有严格理论基础, 能有效减轻SMC-PHD滤波中的粒子退化, 且适用性很强.仿真实验对比了该算法、经典SMC-PHD和基于无迹卡尔曼的SMC-PHD算法的跟踪性能, 验证了该方法无论对势估计还是对目标状态估计的精度都优于其他两种算法.

Multi-target tracking based on improved probability hypothesis density filter

Due to the most recent observational data being unused, the particles in sequential Mote Carlo probability hypothesis density (SMC-PHD) filter which are drawn from prior transition is far away from the real states and may seriously degenerate. Aiming at these problems, we propose a method named square-rooted cubature Kalman sequential Mote Carlo PHD (SCK-SMC-PHD) filter which uses square-rooted cubature Kalman filter to generate the proposal density function and obtains the present particles states by sampling from the proposal density function. The proposed method which can alleviate particles degradation effectively has rigorous mathematical theoretical basis and strong adaptability. Simulation compares the proposed method with C-SMC-PHD filter and the SMC-PHD based on unscented Kalman filter. The results show that the proposed SCK-SMC-PHD filter has a higher accuracy in estimation of both individual state and target number than the two methods mentioned above.