A Novel Fragments-based Similarity Measurement Algorithm for Visual Tracking

Various adaptive appearance models have been proposed to deal with the challenges in tracking objects such as occlusions, illumination changes, background clutter, and pose variation. In this paper, first, we present a novel Fragments-based Similarity Measurement algorithm for object tracking in video sequence. Both the target and the reference are divided by multiple fragments of the same size. Then, we find the similarity of each fragment with the overlapped smaller patches by comparing the average intensity value of the patches. The accuracy of the tracking results can be improved by adjusting the size of the patches. Finally we incorporate the global similarity measurement using two kinds of distances between them. This method encodes the color and the spatial information so that it can track non-rigid objects under complex scene. We use this coarse-to-fine method to get a balance between the accuracy and the computational cost. Extensive experiments are conducted to verify the efficiency and the reliability of our proposed algorithm in the realistic videos.