为解决多目标跟踪中数据关联问题引入神经网络UKF算法和分类信息辅助方法。通过计算分类混淆矩阵确定分类信息似然函数,再用该函数调整只利用运动信息的似然函数,使分类信息有效辅助神经网络联合概率数据关联。在杂波环境中对多个近邻且不同种类目标跟踪,针对联合概率数据关联(JPDA)存在的计算组合爆炸问题,利用Hopfield网络解决TSP问题的思路得到神经网络联合概率数据关联(NJPDA)方法。基于以上关联方法得到的关联概率,仿真表明该算法提高了数据关联的有效性和快速性。

To solve the problem of data association in multiple target tracking system, the UKF neural network and classification-aided algorithm was brought into the algorithms. The classification information likelihood function was defined by class confusion matrix, and this function was used to adjust the measurement likelihood function so that the classification information could effectively aid the probabilistic data association. By using Hopfield network solving TSP problem, NJPDA algorithm avoided the large computational burden in JPDA. Based on the prime data association received from NJPDA, the simulation of multi-target tracking in clutter shows that data association is improved by the proposed technique.