Novel side pose classification model of stretching gestures using three-layer LSTM

Abstract

In recent years, low back pain rehabilitation exercises have been widely performed for spine-related illnesses. To facilitate rehabilitation exercises, pose-based human action recognition technique is used to determine human movement from simple videos. Herein, we propose a new stretching side pose classification system using three-layer long short-term memory (LSTM) that can be used in rehabilitation therapy systems. Four types of rehabilitation treatment exercises are selected: bird dog, cat camel, cobra stretch, and pelvic tilt. Features selected based on the high frequency of use for each exercise resulted in improved classification. Consequently, the recognition rate of the selected feature is 97.50%, as classified by the three-layer LSTM model.