针对基于概率假设密度算法(Probability Hypothesis Density,PHD)的高维纯方位多目标跟踪,提出了新型的PHD算法—新型采样准则的基于无迹变换的粒子PHD算法(Unscented Particle PHD based on New Sampling Rule, NSRUP-PHD).新算法对每个目标设计了基于无迹变换(Unscented Transform,UT)的粒子滤波器,不仅解决了非线性滤波估计的问题,而且还通过高斯混合的方式实现了非高斯噪声估计.此外粒子滤波器提出了一种新型的采样手段,通过基于三阶容积准则(Cubature Rule,CR)的粒子方位选择和概率累加的距离延伸,使得采样粒子遍布整个空间的同时保障了粒子概率分布的问题,提高了粒子使用的效率.仿真结果表明NSRUP-PHD能够实现多目标有效跟踪,相比于传统的算法和伪随机采样,新型滤波器和采样手段可改善跟踪效果.%Aiming at high dimensional bearings-only multi-target tracking,unscented particle PHD based on new sam-pling rule (NSRUP-PHD) is proposed. For NSRUP-PHD,a particle filter based on unscented transform is designed,which successfully fulfills the states estimation of non-linear system,and realizes the noise update by mixture of a series of Gauss parts. Besides,NSRUP-PHD provides a new sampling method to accomplish the particles angular resolution with the three cubature rule (CR),and the distance between the certain particle and the expectation is the upper limit of probability integra-tion,certainly,the particles can be spread through the whole sampling space by obtaining the relative probability and getting the distance,which results in enhancing the particles efficiency. Simulation results on NSRUP-PHD prove that new filter and sampling method are more effective for bearings-only multi-target tracking in comparison with traditional filter and pseudo random sampling.
展开▼
