Bayesian Framework Based Group Target Tracking

Group target tracking is one of the important areas of modern target tracking technology,and it is a special case of multi-target tracking.Under completely indistinguishable,partially distinguishable and fully distinguishable conditions,group target tracking is significantly different from multi-target tracking: based on the traditional data association method,it is necessary to consider the overall movement trend and structure information of the group targets,the interactive characteristics between group and individual,individual and individual,the attributes and dynamic changes of group targets,etc.Therefore,it is urgent to study the modeling and solving methods of the comprehensive utilization group target related features.Based on the background of the application of group target tracking and Bayesian theory,this dissertation theoretically studies the group target tracking method based on Bayesian framework,and fully exploits the group target feature attributes and their relationship at the application level.Feature attributes are conveniently integrated into the Bayesian framework for estimation.The proposed Bayesian framework and algorithms can make a more comprehensive,clear and accurate explanation of group target tracking,which has important theoretical and practical significance for improving environmental awareness.The main contributions of this article are summarized as follows:1.A unified Bayesian framework for group target tracking is constructed.Firstly,for the problem of overall movement trend and shape estimation of dense group targets,a single-layer Bayesian framework is given,which can integrate group targets and environmental attribute variable estimation.Secondly,aiming at the sparse group target structure information and the group and individual interaction characteristics,the joint group-individual estimation Bayesian framework is given,which can integrate the interaction relationship of the targets within the group to improve the tracking quality.This paper also summarized the recent developments in the study and applications of group target shape model and group target interaction model.In addition,because the characteristics of nonlinear non-Gaussian and high system dimensions,this paper also introduces the application method of particle filtering and random finite set method in group target tracking.The differences and connections between these basic models and implementation methods offer a foundation for future research.2.A robust filter is proposed for robust group tracking based on random matrix model.A mixture generative model is defined in cases where the clutter density parameter is integrated.An approximation recursive algorithm based on the variational Bayesian approach,to approximate and learn both joint posterior states and clutter density parameter simultaneously,is derived to process the accumulated measurements.In addition,for group target tracking with heavy-tailed noise in clutter,etc.,an explicit distribution is used to describe the non-Gaussian heavy-tail noise based on the Student's t-distribution,and provides a mechanism that infers the significance of each observations contribution to the expected information.The need of arbitrary decisions is then eliminated,and the robust operation is provided which is less sensitive to extreme observation.The effectiveness and robustness of the proposed algorithm are verified by simulation experiments.3.A Bayesian approach to multiple moving group targets tracking is proposed for estimating the shape approximation of the group targets in addition to their kinematics.Within this approach,the group target extensions are modeled with the random hypersurface models,and a new variant of probabilistic multi-hypothesis tracking is used for modeling assignments of measurements to extended targets.By defining the hierarchical Bayesian model of the random hypersurface model,the shape Fourier coefficients and the scale factor are described using the beta distribution and the Gaussian distribution,respectively.These two parameters have hyper-parameters governing their prior distribution,and the uncertain of the measurement source relative to the shape boundary is more accurately formulated.In addition,an approximate measurement update that arises directly from the analytical techniques of the variational Bayesian method is derived to simultaneously estimate the posterior states iteratively,including the shape and kinematics of each group targets.Therefore,the performance of the proposed algorithm is demonstrated with simulated data and better than the traditional method.4.For the sparse maneuvering group target tracking application in clutter environment,the problem of state estimation for a dynamic system driven by unobserved,correlated inputs is considered based on nonparametric Bayesian model.The proposed algorithm combines the multi-hypotheses tracking method with the hierarchical Dirichlet process-Hidden Markov model.The HDP-HMM is used to learn the unknown time-varying changes including the number of modes and the rate of mode transitions,and the MHT is used to generate and manage hypotheses for multi-target tracking.In addition,this paper present an online learning algorithm based on Rao-Blackwellized particle filter,which reduces the dimension of the system sampling space and improves the particle sampling efficiency.The simulation results are presented to demonstrate the effectiveness and superiority of the proposed approach.