Abstract: A competitive networked robot system based on high-speed vision is built.At the same time,by comparison, selection and integration of a variety of identification methods,a quick and effective target recognition and tracking algorithm for competitive networked robots is designed.Dynamic window technique and the improvement of traditional methods further enhance the system's real-time performance.The algorithm can track the moving target robot in complex background,with good robustness and less computation.Experimental results show that it takes less than 5 ms to process one frame of grey image of the size 640×480 pixels.