A particle filter object tracking based on feature and location fusion

For the object tracking under complex scenes, this paper proposes a tracking algorithm combining both the sample feature and the location. According to differences between the object and its neighboring background in different color subspaces, the suitable color subspaces, which make the object outstanding from the background, are selected and the multiple feature descriptions based on the selected color subspace are produced to represent the object. Thus, under the particle filter framework, the similarity matrix which is based on the sample feature and the location are computed. According to the similarity matrix, the algorithm determine the final object location. The results show the proposed algorithm realizes the robustness of the video object tracking to some extent.