Continuous Detection Method for Abnormal Running Posture Based on Wearable Inertial Sensors

Running is a beneficial form of exercise that promotes cardiovascular health, strengthens muscles, and assists in weight management across all age groups. However, abnormal running posture can cause excessive fatigue and increase the risk of injuries. This study proposes a continuous detection method for abnormal running posture based on wearable inertial measurement units (IMUs), aiming to efficiently classify and recognize abnormal running postures through deep learning models by utilizing raw data collected from wearable inertial sensors. Firstly, a systematic comparison of long short-term memory (LSTM), gated recurrent unit (GRU), and Transformer models shows that the four-layer Transformer model achieved a test accuracy of 85.2%, which is superior to that of LSTM and GRU models in terms of classification accuracy. In addition, the study found that using both the accelerometer and gyroscope signals significantly improves classification performance by comparing different sensor signals. Finally, it was found that a single IMU mounted on the waist or chest can effectively replace a dual-IMU combination without compromising detection accuracy, thus supporting a low-cost, low-burden running posture monitoring solution. This study provides an innovative solution for real-time monitoring of abnormal running posture and provides guidance for deep learning model selection, sensor signal configuration, and mounting placement.