基于神经网络的多机动目标跟踪算法

将神经网络理论用于多机动目标跟踪,解决了联合概率数据关联(JPDA)存在的计算量组合爆炸问题。基于神经网络数据关联(NDA)所得到的最佳关联假设,将其与简化信息融合并行自适应滤波算法(DAF)进行有效结合,在保证量测与目标有效关联的同时,还具备跟踪起始和终结的作用,实现了对多机动目标的状态滤波与预测。仿真结果表明,与传统的交互式多模型联合概率数据关联算法相比,新算法在保证多机动目标的跟踪精度及实时性要求的同时,计算量大大减少。

A Much Faster but Still Good Algorithm for Tracking Multiple Maneuvering Targets

Aim.The traditional IMMJPDA(interactive multiple models joint probabilistic data association) algorithm gives good tracking performance but the computational burden increases explosively with increasing number of maneuvering targets.We now present a novel algorithm NDA-DAF(neural data association-data fusion adaptive filtering),which makes good use of neural network and,we belive,succeeds in preventing explosive growth of computational burden while retaining almost the same tracking performance as IMMJPDA.In the full paper,we explain in detail our NDA-DAF algorithm;in the abstract,we just add some pertinent remarks to listing the two topics of explanation:(A) NDA algorithm;eqs.(2) though(15), needed for explaining NDA,can all be found in the open literature and are compiled by us for convenience;(B) DAF algorithm;Fig.1 in the full paper gives the schematic of the two-filter parallel architecture;eqs.(16) though(22),needed for explaining DAF,can all be found in the open literature and are compiled by us for convenience.Finally we give a numerical example to illustrate the application of our novel NDA-DAF algorithm.Fig.2 in the full paper gives the Utype trajectories of two maneuvering targets.Simulation results for comparing IMMJPDA with NDA-DAF are given in Figs.3~6 in the full paper.These results show preliminarily that the novel algorithm NDA-DAF,which makes good use of neural network,has almost the same performance as the traditional IMMJPDA but with greatly reduced computational burden for tracking multiple maneuvering targets.