Research On Group Structure Modeling And Group Target Tracking

Target tracking has always been a hot field in scientific research.With the continuous development of society,tracking scene becomes increasingly complex,and multi-target tracking technology also gets rapid development and application.The concept of group target tracking was then put forward.In group target tracking,the group target is composed of multiple single targets,which move in the same speed or direction,such as group of UAVs and warship formation.At present,group target tracking is widely used in various fields,such as intelligent monitoring,robot visual navigation,indoor positioning,etc.With the development of research,more and more filtering algorithms based on random finite set(RFS)have emerged,such as probability hypothesis density(PHD)filtering,Poisson MultiBernoulli Mixture(PMBM)filtering,etc.For group target tracking,the problem of splitting within the group and merging among different groups in the process of group evolution is called the modeling and updating of group structure,which has very important significance.It enables us to get the group structure information when tracking.This paper focuses on the group structure model and group target tracking algorithm.The research contents are as follows:Using machine learning to break the limitations of the traditional group structure model,the Kernel Fisher Discriminant Analysis(KFDA)group structure updating model is proposed,and the group structure updating model conforming to the characteristics of group structure is obtained through off-line training.Combined with the box particle probability hypothesis density filtering(BP-PHD)algorithm,the group structure modeling and group target tracking are realized.The simulation results show that the proposed model has better tracking performance and smaller error than the group evolutionary network model.Considering the tracking problem of group target contour,an improved group structure model is proposed,which consists of two parts,the first part,using Gamma Gaussian Inverse Wishart(GGIW)model to estimate the shape of the group of target.In the second part,the KFDA group structure update model is used to solve the inaccurate measurement division.In the case of group cross,the measurement sets generated by different groups will overlap,so that the correct measurement set cannot be obtained,and the target estimation will be missed in tracking.The KFDA group structure update model was used to divide the measurements for the second time,which improved the accuracy of the division and avoided the missing estimate of the target.In PMBM filter,the target state modeling can be divided into poisson and Bernoulli two parts,with the nature of the conjugate prior.The PMBM filter has very good performance in multiple target tracking,therefore this thesis applied it in the group target tracking,under the scenario of coexistence of point and group targets,make it combined with improved group structure model.The experimental results of the point and group target tracking algorithm combined with the improved group structure model and PMBM filter show that it can not only track point and group targets correctly,but also improve the tracking accuracy,and estimate the number of groups more accurately.