高斯粒子PHD滤波的多个弱小目标TBD算法

针对现有多个弱小目标检测前跟踪(track before detect, TBD)算法存在的跟踪精度低,算法复杂度高等问题,提出一种新的基于概率假设密度(probability hypothesis density, PHD)的TBD算法。所提算法通过高斯粒子滤波对PHD中的各高斯项进行递归运算、进行多帧能量累积,并提取高斯项的均值为目标的状态,达到检测与跟踪多个弱小目标的目的。算法在随机集滤波框架下完成未知数目的多个弱小目标跟踪,不仅充分利用粒子滤波的非线性估计能力,同时避免了传统算法利用模糊聚类进行目标状态提取所带来的跟踪精度低等问题。仿真结果表明,所提算法与传统方法相比,在降低算法复杂度的同时,对多个红外弱小目标具有更加良好的实时检测和跟踪性能。

Track before detect using Gaussian particle probability hypothesis density

In order to avoid the low tracking accuracy and high complexity problems in the conventional algorithms, a novel track before detect algorithm based on probability hypothesis density (PHD) filter is proposed for the tracking and detection of the multiple dim targets in the infrared image. With the Gaussian particle filter, the Gaussian components in PHD can be operated recursively and extracted as the states of targets. The algorithm can realize the tracking and detection of the multiple dim targets by the energy accumulation. With the theory of the random finite set, the algorithm performs the multiple dim targets tracking with unknown number. It can not only make use of the nonlinear estimation ability of the particle filter but also avoid the tracking inaccuracy which is brought by the fuzzy clustering. Simulation results with the infrared images show that the proposed algorithm has the low complexity and the better performance in the detection and tracking multiple dim targets than the conventional algorithm.