Abstract: Particle swarm optimized particle filter (PSOPF) can deal with leanness and divergence of the traditional particle filter to some extent. However, robustness and tracking precision need to be improved as the system observation model is limited to the use of observed data. Based on the imaging mechanism of the
forward-looking sonar, extraction of multiple features and fusion of the features with PSOPF are discussed.An optimized algorithm using adaptive weighted eigenvalues as fitness is proposed. It makes particles move to the high likelihood zone by updating their speed and position. Experiments are performed to track underwater targets from sonar image sequences. The results show that the multi-feature fused PSOPF can improve robustness and decrease risk of leanness and divergence. It can achieve high tracking precision with fewer particles. The advantages make it suitable for underwater target tracking.

Key words:  target tracking, particle swarm optimization, particle filter, multi-feature fusion, forward-looking sonar