光电工程, 2009, 36 (3): 22, 网络出版: 2009-10-09
一种新的目标跟踪算法研究
A New Algorithm of Target Tracking
多目标跟踪 集中式融合系统 多传感器 概率假设密度 粒子滤波 multi-target tracking centralized fusion system multi-sensor probability hypothesis density particle filtering
摘要
当采用概率母函数将单传感器PHD 滤波推广到多传感器情形时,针对计算繁琐,难于实现的问题,本文基于集中式融合系统的有序滤波思想,提出多传感器、多目标有序粒子PHD 跟踪算法,该算法通过选取与各传感器相关的重要性密度函数,层层更新各传感器的采样粒子,达到多传感器多目标有序PHD 跟踪。实验结果表明,当仅仅使用单传感器对多目标进行跟踪时,虚警概率较高时一些粒子会严重偏离原始目标轨迹,导致目标数目估计出现偏差,而采用多传感器多目标有序PHD 跟踪可以有效减小多目标距离跟踪误差,提高跟踪精度。
Abstract
Aiming at complicated computing and difficulty of being realized when extending single sensor PHD filtering to the multi-sensor case by means of probability generating function, a multi-sensor multi-target sequential particle-PHD tracking algorithm is proposed based on the thought of sequential filtering for a centralized fusion system. The algorithm chooses the importance density function with regard to every sensor, and updates sample particle of every sensor layer by layer. Finally, the multi-sensor multi-target sequential PHD tracking is realized. Experimental results show, when multi-target is tracked only using single sensor, some particles can deviate true trajectories of target, which causes the error of estimated numbers of targets. However, the multi-sensor multi-target sequential particle-PHD tracking algorithm can reduce distance error and improve tracking accuracy effectively.
齐立峰, 冯新喜, 惠小平, 白剑林. 一种新的目标跟踪算法研究[J]. 光电工程, 2009, 36(3): 22. QI Li-feng, FENG Xin-xi, XI Xiao-ping, BAI Jian-lin. A New Algorithm of Target Tracking[J]. Opto-Electronic Engineering, 2009, 36(3): 22.