Comparative evaluation of triboelectric nanogenerator-based and inertial measurement unit-based systems for smart gait analysis in human activity recognition

In this work, we evaluate the performance of a self-powered sensing unit for human activity recognition (HAR). The system consists of two triboelectric nanogenerators (TENGs) that are embedded inside the insole of the left shoe. An Inertial Measurement Unit (IMU) is also attached to the ankle of the left foot to provide a reference point. The IMU will compare the performance of the TENG-based HAR to that of the IMU-based HAR. Five physical activities were monitored in this study: walking on a flat surface, walking upstairs, walking downstairs, running, and jumping. Each segment of activities was designed with a few seconds of idle time before and after for better annotation and segmentation. The idle periods enhanced data separation, reducing overlap between activities and ensuring clearer, more accurate analysis for each movement type. We observed that TENG data identifies all five distinct activities, and machine learning algorithms classify these activities with sufficient accuracy and minimal data preprocessing. Among the various tested algorithms, the highest performance was obtained within the Random Forest classifier, reaching an accuracy of 93%. This work proves that TENG-based motion sensing is suitable for activity recognition for portable Internet of Things (IoT) devices with lower energy expenditure.