Wearable EMG–POF Multimodal Sensing System for Gait Recognition

Health monitoring for the elderly has become a pressing issue, and gait monitoring using wearable devices is being increasingly recognized as a crucial approach for health surveillance. However, many existing devices are limited by their data collection methods, which often fail to comprehensively capture physiological information, thereby restricting the identification of abnormal gait patterns. In this work, we developed a multimodal sensing system that integrates electromyography (EMG) amplifiers with polymer optical fiber (POF) pressure-sensing insoles. The system captures variations in leg muscle activity and plantar pressure during movement, enabling portable and high-precision gait monitoring. It efficiently and accurately analyzes key parameters within the gait cycle, such as cadence and stride length, achieving an error margin of only 0.578%, thus providing a solid foundation for gait pattern recognition. In experiments focused on daily activity gait recognition, we utilized time–frequency statistical features and a wavelet packet decomposition (WPD) algorithm, combined with machine learning techniques, resulting in a recognition accuracy of 95.068%. Furthermore, to enhance the system’s performance in detecting abnormal gait, a feature-optimized bidirectional long short-term memory (BiLSTM) neural network was employed to train on multimodal data, achieving an accuracy of 98.095%. The experimental results demonstrate that the system can effectively distinguish between routine activities and abnormal gait states, thereby contributing to the advancement of wearable devices for health monitoring in elderly individuals.