Study On Recursive Target Tracking And Classification Methods For Passive Acoustic Sensor Networks

Passive acoustic detection system (PADS) utilizes acoustic sensors to receive bearing-only data for target tracking and classification. Compared with active radar system, it has the advantage of good detecting ability, immune to electromagnetic interference, high battlefield survivability, low cost and all weather operations. Therefore, the PADS is regarded as a useful supplement for the active radar system. However, challenge remains and we are looking forward to tackle with the nuisances of nonlinearity, incomplete observation and signal time delay which make the tracking and classification problems more complicated. In this paper, we studied the problems of targets tracking and classification for the PADS. The main results and achievements are given as follows:(1) Unified mathematical models of target state and class attributes as well as time delay measurements were built for single and multiple targets tracking and classification problem.(2) Bearing-only acoustic data can not be used directly for the effects of signal time delay in the PADS. In order to make the delayed measurements available for tracking, synthesized measurements could be get through posteriorly optimizing the joint measurements which derived from every sensors in different scans using the methods of probabilistic modeling. Then, a TD-PMHT algorithm can be constructed by embedding a Bayesian smoother within Expectation and Maximization (EM) framework for single target tracking problem.(3) An application of the joint tracking and classification algorithm based on mixture particle fiter (MPF-JTC) in the PADS was realized. Furthermore, a new mixture unscented particle joint tracking and classification algorithm (MUPF-JTC) was proposed by adopting the methods of unscented transform (UT). Since better particle proposal distribution can be computed by embedding unscented Kalman filters (UKFs), the MUPF-JTC can achieve better tracking and classification results than a common MPF-JTC algorithm.(4) The data association that was brought by multi-target further complicates the trakcing and classification problems in the PADS. To slove it, a multi-target time delay probabilistic multiple hypothesis tracking algorithm (MTD-PMHT) and another multi-target time delay probabilistic multiple hypothesis joint tracking and classification algorithm ( MTD-PMHT-JTC) were proposed based on the TD-PMHT and MUPF-JTC algorithms as well as the PMHT technology. The MTD-PMHT can tackle the problem of data association. The MTD-PMHT-JTC can accomplish the classification work of multi-target. Simulation results confirm the efficiency of the proposed algorithms.