Objectness-based smoothing stochastic sampling and coherence approximate nearest neighbor for visual tracking

Abstract

In visual tracking, most of the tracking methods suffer from abrupt motions. To address this problem, we propose a novel method for tracking abrupt motions using objectness embedded in smoothing stochastic sampling and improved Tree coherency approximate nearest neighbor. An improved coherence approximate nearest neighbor is utilized to estimate the promising regions as prior knowledge. Moreover, objectness is employed as an objectness proposal function for handling dynamic motions. Finally, both prior knowledge and objectness proposal are integrated into the smoothing stochastic approximate Monte Carlo to predict a new state of the target object. Experimental comparison with other tracking methods and proposed method was carried on some of the challenging video sequences. Experimental results demonstrate that our proposed method outperforms other state-of-the-art tracking methods for dealing with abrupt motions in terms of effectiveness and robustness.