Real-Time Seat-Based Sitting Action Recognition Using Soft Pressure Sensors and Embedded Machine Learning

초록

Seat-based sitting behavior monitoring is an important research area due to its impact on spinal health, work performance, and long-term comfort. However, challenges remain in accuracy, stability, cost, deployment flexibility, and data privacy. This article proposes an integrated system combining soft pressure sensors, embedded machine learning, and a flexible communication architecture. The pressure sensor, fabricated from a conductive textile and an Au film, achieves a sensitivity of up to 0.46 kPa(-1), with response/recovery times of approximately 60/50 ms. The compact machine learning model (4.34 KB) achieves 99.7% accuracy in recognizing six sitting actions, including trunk tilting and sitting transitions. All processing and inference are performed locally on the embedded board, with only the classification results transmitted to the cloud to enhance data security. The system focuses on trunk tilt and sitting transition recognition rather than full-body ergonomic posture assessment. These results demonstrate the feasibility of real-time, resource-efficient monitoring of sitting actions on edge Internet of Things (IoT) devices. This pilot study used data from one participant with a fixed chair and sensor setup in a single recording session. Therefore, the results show feasibility under controlled conditions, not generalizability across subjects. Sensors Edge Al-loT ... 71 r"Or

키워드