Video Sitting Posture Recognition of Human Skeletal Features Based on Deep Learning

Abstract

Sitting in a poor posture for long time can be detrimental to your health. In this regard, video-based detection of poor sitting posture and providing alerts can help people to improve their physical and mental health and productivity. The use of computer vision to detect human sitting posture is a simple method, but there is a problem of low accuracy in practical applications. In this paper, we propose a video sitting detection method based on multidimensional skeletal features of the human body. Using OpenPose to extract human information features from video sequences, global angle information and local angle information formed by human skeletal segments are used as dimensional features, and sitting posture recognition is detected by deep learning with LSTM models. Experiments show that the method effectively improves the accuracy rate.