基于空间直方图的多目标粒子滤波跟踪

针对复杂场景中多目标跟踪问题,本文给出了目标的出现与消失、遮挡等模型描述,将其统一到粒子滤波的框架下,提出了一种可以处理目标数可变的多目标跟踪算法.对场景中的目标数建立马尔科夫模型,采用转移概率矩阵描述跟踪过程中目标出现,消失的情况;在状态表示中增加辅助变量,明确目标之间可能的遮挡;采用目标空间直方图建立基于唯一性原则的观测似然函数,通过后验概率分布估计目标数及目标状态.实验结果表明,本文算法能有效地处理跟踪过程中的目标数变化、目标遮挡等问题,实现多目标的正确跟踪.

Particle Filter Algorithm for Multi-target Tracking Based on Spatial Histogram

For the multi-target tracking problem of complex backgrounds, the models of appearance, disappearance and occlusion of target in observation scene were described, and a probabilistic multi-target tracking algorithm based on particle filter was proposed. The variable number of targets was modeled by a Markov chain, and a hidden variable was augmented in state representation to represent possible occlusion explicitly. An exclusion principle based on observation likelihood was constructed with the spatial histogram of the target, and the state and number of targets were estimated by the posterior probability. The experimental results show that the proposed algorithm is robust to the problems, such as variable target number, the similar appearance disturbance, and short-time occlusion.