基于LSTM子网络聚合的雷达TWS跟踪

Radar TWS Tracking Based on LSTM Sub-networks Aggregation

为了提高雷达TWS跟踪的准确性和时效性,解决在高密度杂波环境下传统跟踪算法性能大幅下降和模型依赖问题,论文提出了一种免模型、纯数据驱动、端到端学习的雷达多目标跟踪框架和基于长短期记忆网络(LSTM)子网络聚合的雷达TWS跟踪方法。该方法将含噪声的量测通过数据关联子网络,在多目标、杂波和漏检情况下可正确输出与已跟踪目标的关联矩阵;并通过轨迹预测子网络预测和更新机动目标的运动轨迹;最后使用航迹管理子网络来管理航迹生命周期。通过仿真实验证明了该方法的可行性,与传统TWS跟踪算法相比,该方法具有更高的跟踪精度和关联正确率,而且不依赖复杂的先验运动模型和杂波分布知识。

Aiming at improving the accuracy and timeliness of radar TWS tracking and solving the problems of model depen⁃dence and the performance degradation of traditional MTT algorithms in a high-density clutter environment,a model-free,da⁃ta-driven and end-to-end multi-target tracking framework and a radar TWS tracking method,based on long short-term memory net⁃work(LSTM)sub-networks aggregation is proposed in this paper.Firstly,this method imports the noisy measurement through the data association sub-network,which can correctly output the association matrix with the existed target in the case of multiple tar⁃gets,clutter and missed detection.Then a trajectory prediction sub-network can predict and update the maneuvering target's state.Finally,the trajectory life cycle is managed through the trajectory management sub-network.This approach is demonstrated over a simulation scenario.Compared with the traditional TWS tracking algorithm,it has higher accuracy of tracking and association,and does not require complicated prior motion model and knowledge of clutter distribution.