基于PHD的多目标检测前跟踪改进方法

An Improved Multitarget Track-Before-Detect Algorithm Based on Probability Hypothesis Density Filter

基于概率假设密度粒子滤波的多目标检测前跟踪方法(PF-PHD-TBD)存在目标数目估计不准确、状态估计精度不高等问题。借鉴Rao-Blackwellised粒子滤波(RBPF)将目标的状态空间进行降维分解,分别采用线性与非线性滤波器进行跟踪的思想,在PF-PHD-TBD的预测与更新过程中采用RBPF方法,以最优卡尔曼滤波对目标速度分量进行处理,以粒子滤波对位置分量进行处理,显著降低了运算复杂度,相比仅使用粒子滤波时过分依赖目标位置信息的缺点,充分利用了位置与速度之间的关联特性,提高了目标数目估计的准确度和状态估计的精度。最后用仿真实验验证了所提方法的有效性。

The particle probability hypothesis density filter based track-before-detect（PF-PHD-TBD） always exhibits poor performance in the estimation of targets＇ number and state. In consideration of the Rao- Blackwellised particle filter （RBPF） usually dividing targets＇ state dimensions and independently estimating the linear/nonlinear state component with linear/nonlinear filters, we apply RBPF in the predicting and updating steps in PF-PHD-TBD to estimate the speed component with optimal Kalman filter and the position com- ponent with particle filter, which apparently reduces the computation complexity, and enhances the accuracy of the estimation of the targets＇ number and precision of states, due to making full use of the correlation characteristics between speed and position. Finally, simulation shows the efficiency of the proposed method.