Improved Probabilistic Multi-Hypothesis Tracker for Multiple Target Tracking With Switching Attribute States

Abstract

The probabilistic multi-hypothesis tracker (PMHT) is an effective multiple target tracking (MTT) method based on the expectation maximization (EM) algorithm. The PMHT only uses the kinematic information to solve the problem of measurement to target association. However, in some applications, other information such as attribute measurements of targets may be available, which has potential to reduce misassociations and improve the tracking performance. Integrating attributes into the PMHT may suffer from the switch of attribute states and the instability of attribute measurements. In this paper, an attribute-aided association structure for the PMHT is proposed to consider the uncertainty in both attribute states and attribute measurements. The attribute characteristics are described by the hidden Markov model (HMM), and the joint probabilistic model of kinematic and attribute properties is derived. The attribute states are estimated by the Viterbi algorithm and the data association is improved by the extracted attribute information. Simulation results show that the proposed algorithm has better performance when the attributes of targets are available.