Multi-Target Tracking Algorithm For Dense Targets Tracking In Clutter

The measurement data which is detected by sensor is handled in multi-target tracking, we filter and optimize the data to find the target, and estimate its motion state. We may encounter all kinds of complicated environment in the process of multi-target tracking, and some problems cannot be solved by existing methods. The dense targets tracking in clutter is one of the problems. In the situation of dense target tracking, the track maybe easily attracted by measurement from other target, it will cause track coalescence phenomenon in parallel neighboring or small-angle crossing scene. In the process of data association, we estimate the clutter density in a limited region(for example, the validation gate) and the measurement from other target would be seen as a clutter, the measurement cause by dense target may increase the estimated value of clutter density, and the high clutter density estimation value can bring difficulty to the tracking. In this paper, we study the two problems in dense target tracking, and improve the performance of avoiding track coalescence in data association, then we estimate the clutter density online, and proposed a new method which is not affected by the target measurement. The main work in this paper is as follows:(1)Avoid track coalescence. In the situation of dense target tracking, the closely spaced target may bring great complexities to the target tracking. When the targets keep parallel neighboring or small-angle crossing, measurement originated from targets can easily fall into the overlapping region of the validation gate, and the neighbor targets all update with this measurement, this may lead to that one track is attracted by the other, and the tracks stay close to each other, and finally coalescence. We study three methods to avoid track coalescence: Exact Nearest-Neighbor Probability Data Association(ENNPDA); Exact Nearest-Neighbor Joint Probability Data Association(ENNJPDA); Scaled Joint Probability Data Association(SJPDA). The basic method to avoid track coalescence of these algorithms is: 1 increase the correlation intensity of the target measurement and its own track; 2 decease the correlation intensity of the target measurement and other track. We analysis the efficiency of avoid track coalescence and the effect to normal track of these three methods. The simulation shows, ENNJPDA has the best performance in tracking closely-spaced targets, and would not affect the normal track.(2)Clutter density estimation method. The normal method to distinguish the clutter and the target measurement is the probability method, and clutter density estimation is one of the important parameters of probability. The early target tracking algorithm is based on data association, but rarely estimate the clutter and the noise of system. An accurate estimationresult of clutter measurement density can provide a significant benefit in target tracking. If we estimate the clutter density higher than the real value, which may lead to the difficulty in confirm the track and separate the target, and the lower clutter density would result in the increasing number of false tracks. In dense target tracking, the closely-spaced target measurements would increase the estimate value of clutter density greatly, however, most of the existing clutter estimate methods have no consideration about the target measurement. In this paper, we study about the uniform assumption estimation and spatial sparsity estimation method, and propose a new unbiased algorithm to handle the closely-spaced target tracking. In the new method we subtract the probability of target existence from the number of measurement we used for clutter density estimation in order to get the unbiased estimation of clutter intensity. The new method avoid the higher estimate of clutter density in closely-spaced target tracking, and get the clutter density which is approach to the real value. The simulation show the better performance of new method than the sparsity estimator.