A stochastic resampling based selective particle filter for visual object tracking

摘要

In this work, a particle filter based algorithm has been proposed for visual object tracking. The key idea of the tracking algorithm is to combine a modified particle filter with bat algorithm to reduce sample degeneracy, sample impoverishments, memory requirement and number of particles and to increase tracking accuracy. The proposed particle filter includes a new resampling algorithm which has been proposed by modifying meta-heuristic bat search algorithm. A four dimensional color histogram based model is used which suppresses background color present inside the foreground template, boosting the foreground histogram, when size of the object shrinks, but the template size remains intact. The motion dynamics model further reduces the chance of sample degeneracy among the particles by adaptively shifting mean of the process noise. The proposed algorithm has been compared with other particle filter based visual object tracking algorithms and has been found to be working satisfactorily.