基于随机集理论的多目标跟踪方法

Multi-target Tracking Based on Random Set Theory

多目标跟踪问题通常包括目标信号的检测与目标状态的估计,同时还涉及到对探测范围内目标数量的确定。传统的跟踪方法将目标检测、状态估计与数量确定分别使用独立的模块或算法来处理。在这种模式下,每个模块仅考虑测量数据中与其功能直接相关的信息,模块之间没有信息的交互,因而很难得到全局最优的解。基于随机集理论的多目标跟踪方法将场景内的全部目标看作一个全局变量,目标状态与目标测量分别构成各自的随机有限集。从而多目标跟踪问题可以放在一个随机集模型下的贝叶斯滤波框架中研究。在每一个滤波周期内,通过对随机集的处理,实时地估计目标的数量、状态与类型,实现多目标的联合检测、跟踪与识别。

The multi-target tracking problem usually concerns the signal detection and state estimation,where the determining of the number of the targets in surveillance range is also involved.The traditional techniques handle this problem using separate modules or algorithms. In this pattern each module just takes the information directly related to its functions into account in measurement data,there is no information interacting existed within modules with each other,so it is difficult to get a global optimal solution. In method of random set based multi-target tracking all of the targets are considered as a global variable in the scenario,the target states vector and target measurement data constitute a random finite set respectively. With this idea,the research of multi-target tracking problem can be put into a unified framework of Bayesian filtering based on random set theory. In each cycle of filtering,the joint detection,tracking and recognition can be achieved by online estimating the parameters of random sets.