Fast Visual Tracking using Motion Saliency in Video
Shan Li, M.C. Lee · 2007
We present a novel tracking method for surveillance videos where the object and camera motion could be irregular. The tracker is mainly based on the mechanism of motion saliency detection, where regions of interest are detected according to the motion saliency of the moving objects. Without estimating global and local motion explicitly, we generate motion saliency by using the rank deficiency of gray scale gradient tensors within the local region neighborhoods of the image. In addition, multiple spatial features of targets are integrated in the system to assist the region-based tracking task. Experiment results on many videos suggest that the proposed method tracks moving targets efficiently even in videos with unstable cameras.