The Local Probabilistic Data Association Algorithm Based On The Adaptive K-Value Means Clustering

When the joint probabilistic data association(JPDA)is used to track dozens of targets,there is too much calculation caused by too many feasible events,the state of the target is not updated in time,and finally the target is easy to get lost.A method called local probabilistic data association based on the adaptive K-value means clustering is presented in this paper.This method reduces the dimension of the validation matrix through the strip sub-validation matrix,thereby greatly reducing the number of feasible events,speeding up the solution of the probability of measurement-target association,and ultimately greatly reducing the time-cost of target state update.For each clustering of the K-means clustering algorithm,the number of clusters needs to be specified.Once the number of clusters is not selected properly,the cost function cannot reach the global minimum.There are also differences in the effect of multiple times of clustering.Besides,the clustering result is not reproducible.The adaptive K-value means clustering algorithm proposed in this paper includes an excellent initial clustering center selection algorithm and an algorithm for selecting the number of clusters.The former is a reliable algorithm for automatically selecting a series of initial mean points.The selected initial mean values are far away from each other,so the possibility of merging class in the subsequent clustering process is much lower than the randomly selected initial mean,and the number of iterations required to reach cluster convergence is lower than the randomly selected initial mean.The latter is an effective algorithm for selecting the optimal number of clusters.It can automatically select the "appropriate" number of clusters to balance the hardware and performance requirements.In addition,the adaptive K-value means clustering algorithm makes the clustering results reproducible.The local probabilistic data association algorithm utilizes the strip sub-validation matrix,greatly reducing the dimension of the validation matrix,significantly improving the speed of target state update,while maintaining high correlation accuracy.Finally,multiple simulation experiments have further verified the superiority of the method.The results of the simulation experiment show that when tracking cluster missiles with 50 sub-warheads,although the local probability data association based on adaptive K-value mean clustering takes 143.74 s,which is 2.68 times that of nearest neighbor data association(NNDA),its data association accuracy rate is up to 92.22%,a value 18.04 percentage points higher than that of NNDA.Besides,the results of the simulation experiment show that when tracking eight cross-flying targets,the data association accuracy rate of the local probabilistic data association algorithm based on adaptive K-value mean clustering is very close to the JPDA,reaching 97.48%,which is only 1.92 percentage points lower than the JPDA.The time cost of the local probability data association algorithm based on adaptive K-value mean clustering is only 31.94% that of the JPDA.All these experiments results show that the state update speed of the local probabilistic data association algorithm based on the adaptive k-value mean clustering is significantly faster than that of the JPDA,and it can track at least 50 targets at the same time while maintaining a high association accuracy rate.