No abstract

No abstract

In order to solve the poor robustness resulted from the pose variances, occlusion or clutter in some challenging videos, we propose a robust particle filter tracking method based on a new similarity measure SPBRLSO (spatio-bin-ratio based on the local structure orientation). First, the object tracked is represented by a histogram of the local structure orientation, and then the observation likelihood function is constructed using a spatial-bin-ratio similarity. Finally, the tracker is formed by the he observation likelihood function in the frame of the traditional particle filter tracking. Many experimental results show that the proposed method is able to robustly track the objects in visible videos and infrared videos from some challenging videos including pose variances, occlusion or clutter. Furthermore, our tracking performance is superior to other tracking methods including the VTD tracker, conventional particle filter tracker and the spatiogram tracker.