Adaptive hybrid likelihood model for visual tracking based on Gaussian particle filter

摘要

Abstract. We present a new scheme based on multiple-cue integrationnfor visual tracking within a Gaussian particle filter framework. The pro-nposed method integrates the color, shape, and texture cues of an objectnto construct a hybrid likelihood model. During the measurement step, thenlikelihood model can be switched adaptively according to environmentalnchanges, which improves the object representation to deal with the com-nplex disturbances, such as appearance changes, partial occlusions, andnsignificant clutter. Moreover, the confidence weights of the cues are ad-njusted online through the estimation using a particle filter, which ensuresnthe tracking accuracy and reliability. Experiments are conducted on sev-neral real video sequences, and the results demonstrate that the pro-nposed method can effectively track objects in complex scenarios. Com-npared with previous similar approaches through some quantitative andnqualitative evaluations, the proposed method performs better in terms ofntracking robustness and precision. © 2010 Society of Photo-Optical Instrumenta-ntion Engineers. u0001DOI: 10.1117/1.3465563