Underwater Visual Tracking Method for AUV Based on PSOPF

PF (Particle Filter) is getting prevailing in visual tracking. Nevertheless, PF may cause particle leanness and divergence easily because the observation equation is restricted by the utilization of the observed data. According to the specificity of the forward-looking sonar images and the demand of underwater visual tracking for AUV (Autonomous Underwater Vehicles), this paper forms a tracking method based on PSOPF (Particle Swarm Optimized Particle Filter) which fuses PSO (Particle Swarm Optimization) into PF. PSO drives the particles to high likelihood region by updating their position and speed. MAD (Mean of Absolute Differences), the typical correlation operator, is used as the fitness. In order to control the optimization, this paper proposes a global fitness to describe the particles´ global distribution using the peak width at half height around the global extremum. Experimental results show that comparing with the traditional PF, the PSOPF has better robust performance against multi-path disturbance, which makes the PSOPF suit for underwater visual tracking for AUV.