关键词: 随机有限集, 概率假设密度, 贝叶斯方法, 目标跟踪, 光学传感器

Abstract: Gaussian-mixture probability hypothesis density(GM-PHD) filter is a suboptimal Bayesian method for multitarget tracking based on random finite Set theory. This article proposes a GM-PHD based multitarget tracking algorithm for the image plane of a scanning optical sensor. By analyzing the scan characteristics, a target dynamic model and an observation model are established respectively for the typical cone scan type and shave scan type. The principle of GM-PHD is introduced. To handle the low efficiency problem of the original GM-PHD in high clutter density circumstances, some improvements are presented by using the gating technology in traditional multitarget tracking methods. Finally, A multitarget scenario is set up, which contains crossing targets, parallel targets and approaching targets, Monte Carlo simulations are used for the two scan types mentioned above, and the results show that the new algorithm is able to adapt to time-varying number of targets, depress clutters strongly, and enhance efficiency by 10 times as compared with the original method.

Key words: random finite set, probability hypothesis density, Bayesian methods, target tracking, optical sensor