Behavior Recognition On Multiple View Dimension

Abstract

This paper proposes a behavior recognition pattern recognition method based on image recognition and applies it to the field of dance training. Dance is a performing art and graceful dance is inseparable from the dancers’ good training mode. However, not everyone could enjoy high-quality dance education. We provide a dance training system based on 2D pose estimation and binocular stereo vision. The system relies on binocular imaging principle and deep learning model instead of wearable sensing devices or depth cameras to capture dancer’s three-dimensional human movement in real time. Meanwhile, the learner’s movements will be analyzed to get the difference between the movements and the standard dance, the goodness of the movements and the corresponding scores which are feedback to the learner in order to help them correct their wrong movements and show a better dance.