Research On PHD Evolution Network Model Based Group Targets Tracking

Multi-target tracking technology has gained continuous development and wide application,meantime some challenges arise in increasingly complex tracking scenarios.Based on this background,group targets tracking is proposed,which breaks the limitation of traditional multi-target tracking and performs well in dealing with the splitting and merging of groups.A few kinds of probability hypothesis density(PHD)group targets tracking algorithms are proposed,combining with evolving network model and the PHD filter.Several different implementations including Gauss mixture,Sequential Monte Carlo and box particle are studied and analyzed in this thesis.Firstly,the basic knowledge of PHD filter is introduced.Then some improvements in group targets tracking aimed at the common implementations are put forward,the novel approach are proposed with the combination of evolving network model and these improvements.Evolving network model provides a better method to deal with the structure of groups and its real-time update,it can ensure the accuracy of the group targets tracking and get the information of the groups.This thesis gives labeled Gauss mixture(GM)PHD group targets tracking algorithm firstly,which retains the advantages of traditional GM-PHD filter,such as the small computational amount and the simple realization method.The algorithm estimates information of groups through evolving network model.It has been proved that the feedback of group structure information in the filtering process is helpful to improve the tracking performance.By comparing the labeled GM-PHD and labeled Gauss mixture cardinalized PHD(GM-CPHD)group targets tracking algorithm,this thesis analyzes the simulation results under a high clutter rate and low detection probability environment,The applicability of this two algorithms are analyzed by combining with the average running time.The box-particle evolution networks PHD filter for group targets tracking is proposed to decrease the computational effort of the sequential Monte Carlo(SMC)PHD filter.Boxparticle PHD(BP-PHD)filter can solve the problems of the uncertain measurements anddecrease the computational cost.The proposed algorithm obtains information of group targets which is combined with BP-PHD and evolving network models,then feedback those information to the filter.Consequently the algorithm realizes the tracking and number estimation of group targets.Comparative experiments show that the algorithm in this thesis is more efficient than SMC-PHD filter and performs well in high clutter environment.