TriboGait: A deep learning enabled triboelectric gait sensor system for human activity recognition and individual identification

Abstract

Accurately and continuously measuring and collecting data on human gait is critical for human activity recognition and individual identification, enabling various applications in smart homes/buildings, including security authentication, personal healthcare, and intelligent automation. Many sensing technologies have been investigated by researchers recently, such as camera-based, laser-based, and mobile approaches, which have limitations in particular sensing situations, such as environments with fewer privacy concerns, line-of-sight, and the use of wearables, etc. On the other hand, the floor with the embedded sensor is stable and robust to different circumstances, enabling non-intrusive gait recognition and human identification. Therefore, a triboelectric nanogenerator (TENG)-based gait sensor system installed on the floor is proposed in this paper. Our approach has many advantages in comparison to the existing gait recognition systems, including low cost, simple fabrication, lightweight, and high durability. The TENG-based sensors can be simply embedded into a smart carpet to discern mechanical motions through electrical signals. Furthermore, a deep learning model, deep residual bidirectional long short-term memory network with dense layers (Residual Dense-BiLSTM), is proposed for multichannel floor-based gait recognition. By utilizing this model to analyze the electrical outputs, our system can accurately detect various human activities and distinguish different individuals’ walking patterns, with a recognition rate over 98% and 97%, respectively. We conclude that the proposed deep learning enabled triboelectric gait sensor system has broad applications in security and healthcare.