Advanced visual tracking system based on 3-D motion model of moving object

This paper presents a visual tracking system which detects a moving object from the sampled scenes and analyzes the 3D trajectory. The tracking system features an overhead camera and a zoom function. Since the object moves in a 3D world and occlusion often occurs, the speed and size of the moving object also change from frame to frame. When an meaningful object is tracked with sufficiently magnified scene by zoom control, the trajectory analyzer detects the background movements and estimates the distance from camera to the object. The correspondence of four background regions in stereo frames is developed by a cross-correlation of the projected vertical edges in each region. A peak-level test evaluates the confidence of the result of background movements, and classifies the regions into spatial pattern groups. Finally, the object position and size are determined by temporal differentiation of edge images, and its 3D trajectory is estimated from the camera motion parameters and the distance of the moving object