Traffic Police Gesture Recognition Based on Openpose and GRU

In the field of human posture recognition, especially in the field of Automatic driving, the posture recognition of specific occupational groups has always been a research direction of great concern. In this paper, a traffic police gesture recognition system based on Openpose skeleton recognition and Gated recurrent unit GRU is proposed. Openpose real-time multi person two-dimensional pose estimation is used to extract the traffic police gesture skeleton and keypoints, create multiple 15 frame video datasets, record 8 main traffic gestures, and extract the positions of 14 main joint points that have a great impact on traffic police gesture detection. The motion of human body is detected by recording the position change of joint points. With the help of Openpose, the system can effectively recognize human joints in complex environment and improve the accuracy of human posture extraction. The Gated recurrent unit (GRU) is introduced to extract the timing characteristics of traffic police gestures by using the obtained keypoints instead of images. The experimental results show that the system can better recognize the traffic police gestures, and the accuracy of the moving traffic police gestures can reach 91.51% under the complex background.