Abstract: An improved particle filter object tracking algorithm was proposed to solve object tracking problems in com- plex scene. hhis paper used united histogram to describe target grayscale and gradient direction features imformation, and designed a self-adaptive features fusion observation model to adapt the changing scene. To solve particles degeneracy problem of basic particle filter, a resampling method based on clustering kernel function smooth was proposed. hhe ex- perimental results based on simulation and the actual scenes show that this algorithm is more adaptable and possesses higher accuracy, can track the moving object in complex scene effectively.

Key words: Object tracking, Features fusion, Particle filter, Resampfing