【作者】 张超；

【导师】 李正周；

【作者基本信息】 重庆大学 ， 信息与通信工程， 2022， 硕士

【摘要】 基于随机有限集(RFS)的多目标跟踪方法以集合为基础,将数据关联问题转换为目标状态的最优解和次优解问题,是近年多目标跟踪方法研究的重要方向。本文针对高斯混合概率假设密度滤波器(Gaussian Mixture Probability Hypothesis Density,GM-PHD)在目标新生场景和目标交汇场景存在的理论缺陷,提出基于航迹态势反馈的概率假设密度多目标跟踪方法,理论推导和试验分析所提出方法对新生目标和交汇目标等复杂场景下多目标跟踪能力。主要研究如下:(1)基于航迹起始的新生目标状态估计方法。根据目标在连续时刻的运动稳定性,运用航迹起始准确估计新生目标的状态。该方法主要采用逻辑航迹起始方法对历史状态进行多帧关联,进而将航迹起始成功起始的目标状态反馈至滤波器,修正滤波器新生强度。通过雷达仿真实验和红外仿真实验分别对所提算法进行验证,同时与GM-PHD和TIB-GM-PHD(Track Initialization Based on GM-PHD)进行对比测试,验证了本文所提算法对场景中新生目标的状态和数目估计更加准确。(2)基于航迹反馈的交汇目标状态估计方法。分析目标航迹及其态势,进而将航迹态势反馈引入到跟踪滤波过程,对于即将发生交汇的目标调整其衍生强度,加强交汇目标的分量权重来提高目标辨识度。利用雷达仿真实验和红外仿真实验进行验证,以OSPA距离和目标基数估计为性能指标与GM-PHD和CPGMPHD(Collaborative Penalized GM-PHD)进行对比测试,验证了本文所提算法交汇目标有更好的跟踪性能。本文在传统GM-PHD的基础上增加了航迹反馈层,通过对场景中的航迹情况进行分析从而反馈指导滤波器的模型参数修正,有效提高滤波器对目标新生场景和目标交汇场景的适应能力。更多还原

【Abstract】 The multi-target tracking(MTT)method based on Random Finite Set(RFS)transforms the data association problem into the optimal solution and sub-optimal solutions of the target states,which brings a new idea to MTT.Based on the research of Gaussian mixture probability hypothesis density(GM-PHD)filter,the solutions for the MTT problems in the birth scenario and intersection scenario are proposed in this thesis.The main research of this thesis is as follows:The birth target estimation method based on trajectory initiation is proposed in this thesis.With the trajectory initiation technology,the state of birth target is accurately estimated according to the stable trajectory of target at successive moments.Firstly,the proposed method stores the low-weight components filtered in the tracking process as prior information.On the basis of this,the birth targets are determined according to trajectory initiation.Secondly,the birth trajectory results are obtained and fed back to the filter to guide the update of the birth intensity.Finally,based on the optimal subpattern assignment(OSPA)and the cardinality,the radar and infrared simulation demonstrate that the proposed method achieves better performance on the state and number of birth targets comparing with GM-PHD,TIB-GM-PHD(Track Initialization Based on GM-PHD).The intersected targets estimation method based on trajectory feedback is proposed in this thesis.In the tracking process,a trajectory situation feedback layer is appended to the GM-PHD,which forms the discrete target states into trajectories and analyzes the trajectory situation at the same time.Meanwhile,by adjusting the spawned intensity of the intersecting targets,the proposed method enhances the component weight to improve the recognition of the intersected targets.On the basis of the Optimal subpattern assignment(OSPA)and cardinality,the radar and infrared simulation demonstrate that the proposed method achieves better tracking performance for the intersected targets comparing with GM-PHD,CPGM-PHD(Collaborative Penalized GM-PHD).In this thesis,the trajectory situation feedback layer is appended to GM-PHD.By analyzing the trajectory situation in the scenario,the model parameters of GM-PHD filter are corrected,which effectively improves the adaptability ability responding to the birth scenario and intersection scenario.更多还原