1 Introduction
Multiple targets tracking (MTT) is a challenging problem in point target tracking scenario. MTT techniques were classified as indirect estimation and direct estimation frameworks [1]. Indirect estimation is to use a data association technique to assign the correct measurement to each single target filter that assign to each target [1]. Three main approaches of data association techniques are global nearest neighbor (GNN) approach, joint probabilistic data association (JPDA) method [2] and multiple hypotheses tracking (MHT) method [3], [4], respectively. The GNN approach assigns the most likely observations to existing tracks, which only works well in the case of widely spaced targets, accurate measurements, and few false alarms in the track gates. The JPDA method allows a track to be updated by a weighted sum of all observations in its gate. Both the GNN approach and the JPDA method increase the Kalman filter track covariance matrix to account for the association uncertainty. However, increasing the Kalman filter covariance matrix can exacerbate the problem since an increased covariance matrix leads to even more false observations in the track gate. Moreover, the JPDA method suffers from a problem that tracks on closely spaced targets will tend to come together [3]. Multiple hypotheses tracking (MHT) is the preferred method for solving the data association problem, and it is a deferred decision logic in which alternative data association hypotheses are formed whenever observation-to-track conflict situations occur. Rather than choosing the best hypothesis or combining the hypotheses as in the JPDA method, MHT propagate the hypotheses into the future in anticipation that subsequent data will resolve the uncertainty. The main drawback of MHT method is that it's exhaustively search over all possible hypotheses can be very expensive. However, track-oriented MHT presented by Kurien [4] maintains the tracks that survived pruning and reforms the hypotheses each scan, it maintains less number of hypotheses and hence has less computational load.