Multi-scale Monte Carlo-Based Tracking Method for Abrupt Motion

Video tracking of abrupt motion is a challenging task in computer vision, especially with abrupt scale change. To deal with the problem efficiently, we proposed a novel tracking algorithm based on Markov Chain Monte Carlo sampling method within Bayesian filtering framework. In our tacking scheme, samples were proposed efficiently using the hybrid model of density grid and distance of sub-regions to deal with changes in not only position but also scale. Meanwhile, we introduced mean-shift method to improve the final state according to states of k nearest neighbors of best estimated particle. Experimental results demonstrated the efficiency and robustness of our algorithm.