Real-time tracking of multiple moving objects using split-and-merge contour models based on crossing detection
Shoichi Araki, Naokazu Yokoya, Haruo Takemura · Systems and Computers in Japan · 1999
A moving contour model is an effective method simultaneously to extract and track an moving object. However, it would be difficult to do so correctly for multiple moving objects if they overlap (occlusion), for example if they cross each other. This paper proposes a method which overcomes this difficulty. The method recognizes an occlusion and splits a self-crossed contour model by using an area term, and at the same time this method merges different models which are mutual-crossed during their extraction or tracking. Note that when overlapped multiple objects are separated later, then the merged model is split. Experiments of extraction and tracking of multiple moving objects (walkers) were carried out by using moving images on a workstation and the usefulness of the method was confirmed. An experimental movable system containing an image-processing DSP with the proposed method was constructed, and its real-time processing has been confirmed. © 1999 Scripta Technica, Syst Comp Jpn, 30(9): 25–33, 1999