Maneuvering Targets Track-Before-Detect Using Multiple-Model Multi-Bernoulli Filtering

Abstract

Target tracking using unthresholded raw data under low signal-to-noise ratio circumstance, also referred to as track-before-detect, is a challenging task, especially for the case with varying target number and uncertain target dynamics. This paper deals with the problem of tracking multiple maneuvering targets using raw image observation. The multi-target state is formulated as random finite set and its posterior distribution is approximated by multi-Bernoulli parameters. Multiple model approach is proposed to accommodate the uncertainty of the possible target dynamics, and sequential Monte Carlo method is presented to implement the multiple-model multi-Bernoulli (MM-MeMBer) filter. The state estimates are obtained by combining the result of mode-dependent filtering for the Bernoulli components with high existence probabilities. Simulation results for multi-target track-before-detect application show the improved performance of the proposed method over MeMBer filters in the single-model fashion under the condition of equivalent computational complexity.