自适应GM-PHD滤波器在多目标追踪的应用

Application of adaptive GM-PHD filters to multi-target tracking

针对PHD滤波器中先验概率初始化时,新生目标出现的位置不确定,且目标强度计算区域必须为整个监测区域,造成算法低效率等问题,将原始算法通过PHD滤波的扩展项在预测步骤与更新步骤对新生目标与存活目标进行区分,再通过每一次扫描得到的量测自适应更新得到目标新生强度,依据量测的驱动来避免对先验概率初始化假设的问题。利用OSPA函数作为算法性能监测标准,利用仿真数据和实测数据对改进的算法进行了验证。结果显示,利用量测来驱动新生目标强度函数,对新生目标与存活目标先进性判断,目标数目估计正确率达到97%,OSPA距离较GM-PHD算法下降50%。

The problem is that when initializing prior probability in a probability hypothesis density （PHD） filter, the position of newborn targets is indefinite, and the targeting computations must search the whole monitoring do- main, noticably decreasing efficiency. This paper expands on the original PHD filter, to adaptively distinguish new- born targets from surviving targets in the prediction and updating steps. It then measures again, updating newborn target strength after each scan, thus avoiding the problem of prior probability parameter initialization. The optimal sub pattern assignment （OSPA） function was used to benchmark the performance of the algorithm, plus both simu- lated and tested data were used for validation. Using such measurements greatly enhances discrimination between newborn targets and surviving targets. The results show that estimation accuracy for the number of targets has in- creased to 97%, while the OSPA distance has decreased 50% than the original GM-PHD algorithm.