Smart Shoe System for Accurate Gait Phase Recognition

Jiachen Wang, Jiakang Wang, Tian Liang, Huanghe Zhang

Abstract: Gait phase detection is crucial in rehabilitation and sports analysis. Although wearable sensor systems are widely employed for gait phase detection, their application is often restricted to a two-phase (stance and swing) model. This paper introduces a novel smart shoe system that integrates an inertial measurement unit (IMU), an ultrasonic sensor, and plantar pressure sensors to accurately recognize four distinct gait phases: initial contact, mid-stance, propulsion, and swing phase. However, systems relying solely on plantar pressure sensors face inherent limitations, as the sensors are prone to degradation and failure under the repetitive mechanical stress of long-term use. By integrating IMU and ultrasonic sensors, our system provides a more reliable and comprehensive approach for gait phase recognition. The plantar pressure data served as the gold standard for annotating the four gait phases. We evaluated support vector machine (SVM), recurrent neural network (RNN), and XGBoost algorithms using 10-fold cross-validation. The RNN model exhibited superior performance, achieving an average accuracy of 87.66 ± 5.80% and an F1-score of 0.8765 ± 0.0579 in the four-phase gait classification task. These findings confirm the potential of this smart shoe system as a promising tool for health monitoring, disease prevention, and personalized rehabilitation training, highlighting its capability for future real-time gait analysis.

External IDs:doi:10.1007/978-981-95-2098-5_45