Abstract: Acquiring the motion information of human hands is a very important research problem in the master-slave grasping system for dexterous hands. Used to be the human-computer interface for the master-slave manipulation system for dexterous hands, a binocular stereo computer vision system was investigated to track and recognize the dynamic moving fingertips of human hands. A corresponding identifying algorithm and the experimental system were proposed. Experiment results show that the longest time is 0.4ms to simultaneously recognize the four targets in two images of left and right ones, and the recognition rate reaches 100 percent when the human hand moves in usually used speed, which illustrate that the presented method meets both the real-time and accurate requirements.