自适应观测识别的多目标分类跟踪算法

Multi-target Classification Tracking Algorithm Based on Adaptive Observation Recognition

在现实多目标跟踪任务中,概率假设密度滤波器性能受杂波干扰明显,较大程度地制约目标跟踪的精度和效率。针对这一问题,在概率假设密度滤波框架下提出一种自适应观测识别的多目标分类跟踪算法。首先,依据各个目标的重要程度,自适应观测识别策略采用自适应门限从各时间步观测中识别源于已存在目标的观测、新出现目标的观测和杂波。然后,目标分类更新策略使用不同目标观测分类更新已存在目标的预测强度和新出现目标的先验强度,消弱了不同类目标观测混用对目标后验强度精度的干扰。仿真目标场景验证了该算法的有效性和稳健性。

In realistic multi-target tracking tasks,the performance of the probability hypothesis density filter is obviously disturbed by clutter,which greatly restricts the accuracy and efficiency of target tracking.To overcome this problem,an adaptive observation recognition-based multi-target classification tracking algorithm is proposed within the framework of the probability hypothesis density filter.Firstly,according to the importance of each object,the adaptive observation recognition strategy uses adaptive threshold to identify the observations from the existing object,the newly emerged object and the clutter from each time step observation.Then,the target classification update strategy uses different target observation classification to update the prediction strength of existing targets and the prior strength of new targets,which reduces the interference of different types of target observation mixed on the posterior intensity accuracy of targets.The effectiveness and robustness of the proposed algorithm are verified by the simulation target scenario.