View article

A high performance automatic detection and tracking (ADT) algorithm for passive sonar data based on a one-dimensional hidden Markov model (HMM) is presented. Building on previous HMM research by Thales Underwater Systems, the algorithm, which is a block processing strategy, features multiple model track initiation using the Forward-Backward (FB) procedure with track continuation via the Viterbi algorithm. The FB procedure is implemented in a multi-pass mode to extract multiple target tracks from the data. The stateconditioned observation model is based on probabilistic data association (PDA) including SNR information and features transition probability models for constant velocity and constant acceleration tracks. The latter gives rise to a novel inhomogeneous Markov model. Unlike previous HMM approaches to passive sonar tracking based on 2-D state spaces, only a 1-D state space is required, significantly reducing computational requirements. The 1-D HMM tracker is able to function in a high false alarm density environment typical of passive submarine sonar systems, yielding a very low false track rate while minimising track initiation delay and exhibiting very good track continuity even on high rate of change tracks. Three different simulation scenarios are presented that demonstrate the effectiveness of the technique for both narrowband (NB) frequency and broadband (BB) bearing tracking compared with two conventional Kalman filter based approaches (nearest and strongest neighbour). Since two of the scenarios have been used previously in the public domain, direct comparison of our results is possible. Furthermore, to …