A Multi-Swarm Optimization Using Diverse Charged Particles for Visual Tracking via Template Matching

Abstract

The visual tracking problem can be solved by performing template matching on each frame, which can be treated as a dynamic optimization problem. On the other hand, Charged Particle Swarm Optimization (CPSO) is a PSO variant for dynamic optimization, which maintains swarm diversity by introducing charged particles and repulsion vectors into the velocity update step. CPSO uses a unified pre-determined amount of charge for all particles, which is difficult to deal with various moves of the optimal solution. Instead of a unified value, in this paper, we propose introducing diverse charged particles into CPSO, referred to as Multi-swarm CPSO (MCPSO). Specifically, MCPSO employs multiple sub-swarms and assigns different amounts of charge to particles in each sub-swarm, which leads to diverse moving characteristics of particles. Moreover, each particle is limited to receiving repulsion force only from other particles in the same sub-swarm, which results in various sub-swarms with different behaviors. The optimization performance of MCPSO is assessed by a numerical experiment using five benchmark functions compared with standard PSO and CPSO. In addition, the effectiveness of MCPSO for visual tracking using template matching is verified by a comparison experiment with nine synthetic sequences.