オプティカルフローとエッジを用いた三次元運動物体の複数平面近似による抽出と追跡

抄録

This paper describes a method of tracking a moving object in a 3-D space in a cluttered background by integrating optical flow and edges. The flow field is segmented into regions so that the motion of each region may be approximated by a plane motion. An object region is determined as a set of the plane motion regions with similar 3-D motions (translation and rotation) . The object is tracked by merging and separating the plane motion regions at each frame. In order to estimate the plane motion correctly, the contour of the plane motion region must be obtained precisely. Because the motion boundary of the region is not precise, edges near the motion boundary are used to represent the part of the contour. In order to represent the whole contour even in a cluttered background, the edges are accumulated over frames. If the motion boundary at the overlapping part of the predicted regions can not be extracted, the contour is determined by the edges on the predicted contour. Moving objects are tracked even if the appearances of the objects change drastically. Experimental results for synthetic and real images show the effectiveness of the method.