Robust visual tracking based on adaptive depth-color-cue integration using range sensor

摘要

In this paper, we propose a novel motion detection method for moving human detection based only on range sensor. Different from the state-of-art frameworks on human detection field, our innovation is taking advantages of range data in motion detection and introducing a effective motion capture approach. In preliminary segmentation term the concept of Positive Motion is presented which can obtain points exactly locating on moving objects. Based on this, seed points for region growing can be selected from them to obtain candidate connected area of moving object. During subsequent process, based on couple of criterion, wrongly segmented objects are removed and a multi-object matching strategy is presented for associating objects in consecutive frames which naturally achieves the purpose of real-time multi-object tracking. Experiments corresponding to each section are conducted in various cluttered scenes which validate the effectiveness and robustness of the method. When there are several moving objects in the scene, the moving person can be detected from other nonhuman objects or stationary objects. When there are several moving person moves in the scene, multiple trackers works well in the three dimensional scene.