Research On Key Techniques Of Multi-target Tracking And Data Association Algorithm

With the development of modern warfare weapons and the increasing complexity of modern warfare environments,multi-target tracking technology has received widespread attention.Among them,the data association algorithm is the core of the multi-target tracking problem,and the result of the data association will directly affect the performance of the multi-target tracking.This paper mainly studies the key technology of multi-target tracking,systematically studies the problems of track initiation and data association,and proposes some effective methods.The specific work is as follows:1.The track initiation and data association algorithm are studied.Firstly,the principles of classic track initiation and data association algorithms are analyzed,and then the application scenarios of classic data association algorithms and the performance comparison between the algorithms are analyzed through simulation.Provide a theoretical basis for subsequent research.2.With the increase in the number of targets and the number of echoes,the computational cost of the Joint Probabilistic Data Association(JPDA)algorithm increases exponentially,and many improved JPDA algorithms are proposed for this problem.The Simplified Joint Probability Data Association(SJPDA)algorithm reduces the number of JPDA algorithms by approximately calculating the cluster probability matrix to reduce the number of joint events;The improved Joint Probabilistic Data Association(IJPDA)algorithm updates the state of the target by modifying the probability value of the common measurement.Both of these algorithms reduce the amount of calculation while ensuring a certain tracking accuracy.3.Another method to reduce the calculation amount of the JPDA algorithm is a data association algorithm based on clustering,which uses the clustering algorithm to classify the measurement,avoiding the complicated matrix splitting of the JPDA algorithm.The data association algorithm based on fuzzy c-means(FCM)clustering is to calculate the membership of the measured value from the target,and use this as the weight to update the state;The data association algorithm based on Gaussian mixture model(GMM)clustering is to equivalently measure the measured value to Gaussian distribution,and then calculate the posterior probability of the measured value from the target.After iterative update,select the largest posterior probability for state update.Multi-target tracking data association is a classification problem of targets and observations,and this is the advantage of clustering.Simulation analysis of these two algorithms has achieved very good results.