基于概率假设密度滤波法的3D目标位置跟踪

3D target position tracking based on probability hypothesis density filtering

摘要:

为了实现3D目标位置跟踪，提出了一种基于概率假设密度(PHD)滤波的跟踪方案。方案由2个阶段构成：单视图跟踪阶段和多摄像机融合阶段。单视图跟踪阶段，在时刻k每个摄像机上得到颜色观测值，采用高斯混合概率假设密度(GMPHD)滤波器估计出2D目标位置；多摄像机融合阶段，将得到的目标的2D估计值集合作为数据融合阶段的观测值，并通过GMPHD滤波器估计出目标的3D位置，从而避免观测值与目标状态之间的数据关联。仿真实验结果表明，提出的跟踪方案不但能够可靠地跟踪3D目标位置，而且能够解决在每个摄像机处目标的遮挡问题。

Abstract:

A tracking scheme based on Probability Hypothetical Density(PHD) filtering is proposed in order to realize the 3D target positions tracking. The scheme consists of two phases, namely single view tracking phase and multi-camera fusion phase. In the former, at each camera at time k, the color observations are obtained and then the Gaussian Mixture Probability Hypothesis Density(GMPHD) filter is utilized to estimate the 2D target locations.In the latter, the set of 2D estimations of targets, which is obtained in the former, is considered as observations for the data fusion phase to estimate the 3D target locations by the GMPHD filter, so as to avoid the data association between observations and states of targets. The experimental results show that the proposed tracking scheme can not only reliably track the 3D target positions,but also handle the shielding problem for targets at each camera.