An Advanced On-Line Visual Tracking System

A robot system equipped with an overhead CCD camera which is able to recognize, track and zoom an arbitrary 3-D object traveling at unknown velocities has been built. The problem of on-line visual tracking is formulated as a problem of combining sensor control with computer vision. For image processing, a motion and shape derivation based on eigenvalue problem in object covariance matrices is proposed to estimate an unknown moving object with low computational cost as well as noise detection. The object size, a major shape information for zoom control, is evaluated from the optimized maximum eigenvalues. In the sensor control part, an organic path planning of a pipe-line process between image processing and camera motion control is designed by using the prediction of position error and the feedback of camera motion. The performance of the proposed algorithm has been tested on Sun SS2, a controllable zoom-lens and RV-M1 robot arm, and the experimental results show that the proposed system is valid for numerous kind of moving object with 1.60sec system sampling period.