According to the need for visual target tracking task of the humanoid robot, a new kind of humanoid robot head with high precision and good dexterity was put forward, and the visual perception module and the three degree of freedom robot mechanism were constructed. In order to evaluate the motion ability of humanoid robot head for the target tracking task，the kinematics model of the robot head was established by using the virtual link method, which made the problem of object tracking transform to the problem of inverse kinematics. And the kinematics characteristics was studied by the calculation of condition number and manipulability, which showed that the dexterity of the humanoid robot head was good for the tracking task. Based on this, by using gradient projection method, the synchronization optimization solution was obtained for the inverse kinematics and the avoidance of joint limits. Because of the complexity of image processing, the speed of the visual measurement system was generally slow, and the bandwidth was usually only about 10Hz. Therefore, the target position can not be updated in the measurement blind area between the two measuring points, which led to great tracking error. According to this problem, a prediction method of trajectory in the two visual measurement interval to estimate the target position was proposed. The simulation and experimental results showed that the tracking accuracy can be improved by about 80% by using the trajectory prediction method in the blind area of vision measurement, and the high precision target tracking can be realized. The humanoid robot head and motion control method proposed had good tracking performance for dynamic target.