Kinematical characteristics of human gaits were studied with the aim to achieve meddle-sized league soccer robots' human gait recognition at the non-positive view. The feature quantities for effective representing of guits were adopted and the threshold and frame difference were used to determine the state of characteristic quantities to form the basic method to solve the problem. The three-dimensional information of human bone points was captured by using Kinect to measure the robot view field with the human skeleton model generated by Kinect. By analyzing the human gait situation at the non-positive view, a Kinect- based identification method for gait recognition under the non-positive view was presented. The proposed method can show the azimuth angle of the human body. The practi- cality and effectiveness of the method were verified by experiment, and the results showed that it had a high recog- nition rate at several different non-positive perspectives. The study provides technical reserves for future football match between human and robots.