针对模型驱动方法表达能力有限而难以及时应对目标机动的问题,提出一种基于动态关联波门和深度长短时记忆(DLSTM)网络的机动多目标智能跟踪(MTIT)算法,简记为MTIT-DLSTM-M算法。该算法在MTIT-DLSTM 算法的基础上,首先利用新息定义关联波门以提取有效量测,再采用全局最近邻算法实现数据关联。然后,利用卡方分布理论设计出动态关联波门和机动检测策略,当判定目标出现机动后,清除累积的历史状态和历史量测,重启该目标的跟踪算法流程。仿真实验证明,与现有常规机动目标跟踪算法相比,所提MTIT-DLSTM-M算法运行效率高,且具有更好的鲁棒性,应用前景良好。

Aiming at the problem that model driven methods have limited expressive power and are difficult to deal with target maneuvering in a timely manner, a maneuvering multi-target intelligent tracking (MTIT) algorithm based on dynamic correlation gate and deep long short term memory (DLSTM) network is proposed, which is abbreviated as the MTIT-DLSTM-M algorithm. Based on the previously proposed MTIT-DLSTM algorithm, this algorithm first uses innovation to define an association gate to extract effective measurements, and then uses the global nearest neighbor algorithm to achieve data association. Then, using the Chi-square distribution theory, a dynamic correlation gate and maneuver detection strategy are designed. When it is determined that the target has maneuver, the accumulated historical states and historical measurements are cleared, and the tracking algorithm process for the target is restarted. Simulation experiments show that the proposed MTIT-DLSTM-M algorithm has higher operational efficiency, better robustness, and good application prospects, compared to existing conventional maneuvering target tracking algorithms.

TN957. 52