Non-parametric Multi-target Tracking With Doppler Measurements

In an environment of heavy clutter, multi-target tracking system is a challengingtopic. The existence of clutter measurements often leads the missing or loss of realtargets track, and may result in the emergence of a large number of false tracks. Thisgreatly reduces the credibility of intelligence information and causes great distress forthe operator and the command staff making judgment of the battlefield situation andthreat. Through introducing the extra Doppler dimension measurement probably got byradar into the multi-target tracking, the paper improves the accuracy of informationextraction and distinction. Moreover for the issue of unknown of clutter intensity in theactual system, the paper gives a research of online clutter density estimation, anenhanced the performance of multi-target tracking in heavy clutter. The main work is asfollows:(1) JIPDA tracing algorithm with Doppler (D-JIPDA). The introduction of theDoppler into the joint integrated probabilistic data association algorithm (JIPDA),excludes the clutters in the preselected gate of tracks, and improves the distinctionbetween real targets and false tracks and the accuracy of target state estimation.Simulation results show that D-JIPDA provides a substantial reduction in the number ofconfirmed false tracks and a high estimation Accuracy. For maneuvering targets, thisarticle combines IMM algorithm with D-JIPDA algorithm to process measurement data,and also gets a better tracking performance.(2) Linear multi-target IPDA. In an environment of heavy clutter and dense targets,the computational demand of the optimal JIPDA grows as in a way of "combinatorialexplosion" and is not suitable for large capacity target tracking and real-time system.This paper studies a suboptimal multi-target tracking algorithm framework-linearmulti-target tracking (LM), and based on integrated probabilistic data associationachieves LM-IPDA multi-target tracking algorithm. The basic idea of linear multi-targettracking method is: the measurement produced by the target followed by other track isalso seen as clutter, and then its clutter density is modified, and finally embedded into asingle target tracking methods to achieve the transition of single-target tracking methodto multi-target tracking methods. LM tracking method with a lower computationalrequirement can achieve the same tracking performance as the optimal multi-targettracking.(3) Online clutter density estimation. In most multi-target tracking algorithms, theclutter density is an essential parameter of the data association. The accuracy of clutterdensity, is directly related to the merits of tracking performance. However, in practicalsystems, clutter density cannot be known in advance, and are often non-uniform and time-varying. Target tracking with online clutter density estimation is callednon-parametric tracking. In this paper, three kinds of clutter density estimation methods:uniform assumption estimation, clutter map and spatial sparse estimation are introduced,and are combined with the LM-IPDA algorithm to achieve the appropriatenon-parametric tracking method. Finally, because spatial sparsity estimation methoddoes not consider the contribution of target observation to clutter density estimation, acorrected method using the nearest neighbor independent point to calculate the sparsityis proposed. Experiments show that corrected algorithm for neighboring target trackingperformance achieved significant performance improvements.