Meanshift blob tracking with target model adaptive update

An adaptive model update mechanism for mean shift tracking is proposed in this paper. Gaussian Model always has been used in background model estimation to realize object tracking. It is novel that each bins of kernel histogram is modeled as mixture of Gaussian and the on-line approximation used to update the model. Gaussian distributions are ordered based on the fitness value of weight and covariance. Object model is determined by each Gaussian distributions and weight. Therefore the improved mean shift can not only update object model in time but also deal with object appearance changes and occlusion. Experiments demonstrate that the improved method can track objects under the changes of appearance and occlusion with satisfactory results.