I. Introduction
Multi-target tracking (MTT) has been generally considered the generalization method to estimate and/or update the state of targets, which has been widely used in radar-centric systems where the number of targets are unknown and time-varying. The key to successful MTT lies in finding the effective solution of the data association problem. Generally, there are four main categories, i.e., global nearest neighbor (GNN) [3], joint probabilistic data association (JPDA) [1], multiple hypothesis tracking (MHT) [2], and probability hypothesis density (PHD) [4], to deal with the problem. GNN is the simplest way to solve the problem under the assumption that a measurement is assigned to less than or equal to one target and each target has at most one measurements., which can be used in case of few targets with small measurements. Differing from GNN, JPDA achieves the information of data association by calculating a probability of measurements within the gate of the related target. Both GNN and JPDA are based on single scan to account for the association uncertainty. Both MHT and PHD obtain the data association information by the use of multiple scans, while MHT is a vector-type way to obtain the association uncertainty by choosing the most potential hypotheses and PHD is a set-type approach by estimating the number and position of targets in a measurement set avoiding a prior knowledge of targets.