Deep Learning-Based Recognition of Human Training States During Sit-to-Stand Transfer

The monitoring methods used in traditional lower extremity rehabilitation require complex preparation of the user, which is not only tedious but also prone to secondary injuries. To address this problem, this paper establishes a deep learning algorithm for image recognition to classify and recognize the state features of the human body in the process of sit-stand transfer in the motion phase, firstly, the state images are preprocessed to improve the computing efficiency; then the data are broadened to improve the accuracy of recognition; then the image recognition is used to extract features and train the classification for each motion phase state image. Finally, the results obtained by using the optical motion capture system as a control test verified that the image recognition accuracy can reach 99.5%, meeting the purpose of monitoring the human condition during training.