A system for video tracking of a moving object by the robot-held camera is presented, and efficient tracking methods are proposed. We describe our initial implementation of a system which is capable of tracking a single moving object against highly textured background. A pyramid-based image processor, PVM-1 is employed to support some fast algorithms in locating the moving object from the difference image. Object tracking is accomplished in the static look-and-move mode by the translational motion of a CCD camera mounted on the robot arm. Discussion is given on the implementation of tracking filters and on the effective utilization of multi-resolution processing for the object searching. Finally a method for dynamic look-and-move tracking is proposed for the future improvement of tracking performance.