Posture Recognition for Human-robot Interaction Based on High Speed Camera

Abstract

Physical movements are an important part of people’s activities and communication. The movements of the hands and upper limbs are relatively fast, and the postures in the lower part of the ordinary camera will produce smears, and the images will be clearly obtained only when they are stopped, which seriously affects the recognition speed. This article takes the hands-guessing game in the communication robot as an example. Using a high-speed camera, the corresponding posture can be recognized when the posture is not completed during the hands-guessing game. It is expected that postures can be recognized in a more timely manner in human posture recognition, and the sense of delay in communication can be reduced. The corresponding data is collected and the model is generated after deep learning. In the fast-moving stage of postures, using a ordinary camera, the percentage of time that different postures are unrecognizable is between 50% and 60%. Compared with ordinary cameras, using high-speed cameras, the unrecognizable time percentage of different postures is reduced from 50%-60% to 0%, and the effect is obvious. In human-computer interaction, using ordinary cameras to infer postures, the people participating in the test have a significant sense of delay. With a high-speed camera, this feeling of delay is barely noticeable.