Deep Learning Driven Human Posture Location in Physical Education Teaching

The study of human posture is widely applied in physical education teaching, human motion recognition, and other aspects. With the rise of online teaching, the lack of convenient physical education teaching methods has been able to improve. However, due to the complex structure of human body, the study of human posture is a hard problem of consciousness problem in the area of computer vision. This article mainly studies human posture research algorithms based on deep learning. It uses 101-layer network of ResNet to detect the key points of human body in the image and obtains the categories and coordinates of these key points. In this article, a 101-layer network of ResNet model is constructed to fully learn the visual features of key points in human posture. Secondly, the key point location loss function is improved, and the human posture research is realized by using huber loss function instead of mean square error (MSE) loss function. Finally, experimental analysis shows that compared to traditional integral pose regression (IPR) and location adaptive integral pose regression (LAIPR), the use of ResNet based human posture estimation method for human posture recognition improves precision. It has practical significance for physical education teaching applications.