A hybrid Particle Swarm Optimization with cooperative method for multi-object tracking
Zheng Zhang, Hock Soon Seah, Jixiang Sun · 2012
In the fields of computer vision, multiple object tracking is an active research area. It is a challenging problem mainly due to the frequent occlusions and interactions that happen between the multiple targets. We formulate the multiple interaction problem as an optimization problem and explore Particle Swarm Optimization (PSO) algorithm for the optimal solution. To tackle the problem of premature convergence, we present a new hybrid PSO that incorporates a differential evolution mutation operation with a Gaussian based PSO. Furthermore, by exploiting the specific structure of multiple object interactions, we introduce a cooperative strategy into the proposed PSO for more efficient searching and for conquering the curse of dimensionality. With patch-based observation models, our method can robustly handle significant occlusions and interactions.