基于Renyi信息增量的异质多传感器协同跟踪技术研究

发布时间：2018-04-22 20:49

本文选题：协同跟踪 + 数据融合　；参考：《西南交通大学》2017年硕士论文

【摘要】：在军事、交通、工业等诸多领域,多传感器协同跟踪技术的应用十分广泛。多传感器协同跟踪的目标是最优化多传感器系统的整体跟踪性能。其技术基础为传感器管理技术,通过建立某种传感器管理模型,在各观测时刻实时地为各目标分配最优的传感器组合,实现对监视范围内各目标的跟踪。与同质传感器相比,异质传感器在应用场景和特性上能优势互补,可以提高跟踪的性能。本文的主要工作如下:首先,查阅了大量相关文献,从三个方面综述了目前异质多传感器协同跟踪问题的研究现状。其次,非线性滤波问题和机动目标协同跟踪的精度和稳定性紧密相关。在其相关研究中,DMCKF算法采用协方差矩阵的对角化变换,取代标准CKF中的Cholesky分解,获得算术平方根矩阵,提高了计算的准确度。但是DMCKF和标准CKF在滤波过程中其协方差矩阵有时会失去正定性,导致滤波中断。基于求解各观测时刻协方差矩阵的最邻近半正定矩阵,提出了一种改进的DMCKF算法,确保了滤波过程中观测值容积点的传播不被中断,提升了 DMCKF算法的稳定性。同时,基于改进DMCKF算法,仿真分析了在集中式量测融合和分布式状态两种融合架构下的异质多传感器数据融合算法的性能和适用的情形。然后,针对异质多传感器管理的关键问题:异质多传感器-多目标协同分配问题,提出了一种基于Renyi信息增量的异质多传感器管理算法。该算法通过改进的DMCKF的滤波协方差计算Renyi信息增量,基于求得的Renyi信息增量构造异质多传感器管理模型,在各观测时刻对各机动目标进行异质传感器组合的实时分配。接着,结合改进的DMCKF算法、基于改进DMCKF算法的异质多传感器数据融合算法和异质多传感器管理算法,提出了一种异质多传感器多机动目标的协同跟踪方法。根据异质多传感器的资源分配结果,采用基于改进的DMCKF的异质多传感器数据融合算法获得融合观测值,并在交互式多模型算法(IMM)框架下采用改进的DMCKF对多机动目标进行跟踪。对标准CKF和UKF算法样做了对协方差矩阵求最邻近半正定矩阵处理的改进,仿真验证了改进的DMCKF相比前两者具有更高的协同跟踪精度。同时,改进的DMCKF、CKF和UKF算法相比改进前稳定性显著提升。最后,总结了本文所做的工作,指出了当前研究的不足和下一步研究的方向。
[Abstract]:In military, traffic, industry and many other fields, multi-sensor cooperative tracking technology is widely used. The goal of multi-sensor cooperative tracking is to optimize the overall tracking performance of multi-sensor systems. Its technical foundation is sensor management technology. By establishing a kind of sensor management model, the optimal sensor combination is allocated to each target in real time at each observation time, and the tracking of each target in the surveillance range can be realized. Compared with homogeneous sensors, heterogeneous sensors can complement each other in application scenarios and characteristics, and can improve tracking performance. The main work of this paper is as follows: firstly, a large number of related literatures are reviewed, and the current research status of heterogeneous multi-sensor cooperative tracking is reviewed from three aspects. Secondly, nonlinear filtering is closely related to the accuracy and stability of maneuvering target tracking. In this paper, the diagonalization transformation of covariance matrix is used to replace the Cholesky decomposition in standard CKF, and the arithmetic square root matrix is obtained, which improves the accuracy of calculation. However, the covariance matrix of DMCKF and standard CKF sometimes lose the positive definiteness in the filtering process, which leads to the interruption of filtering. Based on the nearest positive semidefinite matrix of covariance matrix at each observation time, an improved DMCKF algorithm is proposed, which ensures that the propagation of the volume point of the observed value is not interrupted during the filtering process, and improves the stability of the DMCKF algorithm. At the same time, based on the improved DMCKF algorithm, the performance and application of heterogeneous multi-sensor data fusion algorithm based on centralized measurement fusion and distributed state fusion are analyzed. Then, a heterogeneous multi-sensor management algorithm based on Renyi information increment is proposed to solve the key problem of heterogeneous multi-sensor management: heterogeneous multi-sensor multi-objective co-assignment. The algorithm calculates the increment of Renyi information through the filter covariance of improved DMCKF, and constructs a heterogeneous multi-sensor management model based on the obtained increment of Renyi information, and realtime allocates the heterogeneous sensor combinations to each maneuvering target at each observation time. Then, based on the improved DMCKF algorithm, the heterogeneous multi-sensor data fusion algorithm and the heterogeneous multi-sensor management algorithm based on the improved DMCKF algorithm, a heterogeneous multi-sensor multi-maneuvering target cooperative tracking method is proposed. According to the resource allocation results of heterogeneous multi-sensor, the heterogeneous multi-sensor data fusion algorithm based on improved DMCKF is used to obtain the fusion observations, and the improved DMCKF is used to track multiple maneuvering targets under the framework of interactive multi-model algorithm (IMM). The standard CKF and UKF algorithms are improved to deal with the covariance matrix to find the nearest positive semidefinite matrix. The simulation results show that the improved DMCKF has higher tracking accuracy than the former two algorithms. At the same time, the stability of the improved DMCKF / CKF algorithm is improved significantly compared with that of the UKF algorithm. Finally, this paper summarizes the work done in this paper, points out the shortcomings of the current research and the direction of the next research.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212

【参考文献】