Research On Resolvable Group Targets Tracking Algorithm Based On Labeled Random Finite Set

A group target is composed of multiple sub-targets that have a collaborative relationship.According to the resolution of the sensor and the structure composition of the group targets,the group targets can be divided into resolvable group and unresolvable group.In recent years,with the improvement of sensor resolution,resolvable group targets tracking has become a hot topic of widespread attention both domestically and internationally,but in general,existing tracking algorithms have the following problems:First,when tracking large batches of group targets,the computational complexity of the existing methods increases exponentially,and also increases the difficulty of group structure estimation;second,the structural parameters of resolvable groups are unknown,which increases the difficulty of state estimation of group targets;third,merging/splitting behaviors may occur between different subgroups,which makes the collaborative relationship between group members complicated and difficult to estimate.Aiming at the above three problems,this thesis is based on the framework of labeled random finite sets(LRFS),and the main research components include:I.Research on large-batch and multi-structure group targets tracking algorithm based on serial generalized labeled multi-bernoulli(GLMB)filterAiming at the problem that the computational complexity of existing methods increases exponentially when tracking large batches of group targets,a large-batch and multi-structure group targets tracking algorithm based on serial GLMB filter is researched.The algorithm first introduces graph theory to model the relationship between resolvable group targets,and uses thresholding method and adjacency tables to divide the large-batch and multi-structure group targets into multiple subgroups,then tracks the corresponding subgroups targets sequentially and serially,and finally merged to obtain the tracking results of the large-batch and multi-structure group targets.In addition,an average optimal sub-pattern assignment(OSPA)metric is given in order to improve the computational effectiveness of the OS PA metric.Simulation experiments show that the serial GLMB algorithm not only maintains a high estimation accuracy,but also spends less calculation time.Ⅱ.Research on joint estimation algorithm of group parameters and target states with unknown deviation vectorsAiming at the problem of resolvable group targets tracking under the condition of unknown structural parameter deviation vectors,a two-stage estimation method based on the combination of GLMB filter and recursive least squares(RLS)algorithm is researched.In the first stage,K-Means clustering is used to obtain equivalent measurements of the subgroups,and the GLMB filter is used to estimate the states of the centroids of the subgroups;in the second stage,the RLS algorithm is used to estimate the deviation vectors between the group targets and the states of the targets.Additionally,the inner product of velocity and position deviation vectors between group targets is used to estimate the group structures;and the number of eigenvalues of the adjacency matrix is used to estimate the number of subgroups.The simulation experiment shows that the OSPA error of the two-stage algorithm is smaller than that of the GLMB algorithm,the performance is more stable,and the number of groups can also be estimated.Ⅲ.Research on merging/splitting tracking algorithm of resolvable group targetsAiming at the problem of resolvable group targets tracking under merging/splitting conditions,a resolvable group targets merging/splitting tracking algorithm based on iterative self-organizing data analysis(ISODATA)method and GLMB filter is researched.A criterion for judging the merging and splitting of group targets is established:judging whether splitting behaviors occur according to the distance between group members and subgroup centroids;and judging whether merging behaviors occur according to the distance between subgroup centroids.The method is able to dynamically estimate the states,number,subgroup structures and number of subgroups of resolvable group targets.Simulation experiments show that the algorithm can well estimate the states,number,internal structure and subgroups number of resolvable group targets in group merge/split scenarios.In summary,based on the L-RFS framework,this thesis aims at three problems:large-batch and multi-structure group target tracking problem,resolvable group targets tracking problem under the condition of unknown structure parameter deviation vectors,and the resolvable group targets tracking problem under the condition of merge/split,three algorithms are researched respectively:a largebatch multi-structure group targets tracking algorithm based on serial GLMB filter,a joint estimation algorithm of group parameters and target states under the condition of unknown deviation vectors,and a resolvable group targets merging/splitting tracking algorithm based on ISODATA algorithm and GLMB filter.The simulation results verify the feasibility and effectiveness of the algorithms proposed in this thesis.The research results of this thesis have positive theoretical research significance and practical application value.