Extracting Muscle Geometrical Features With a Fabric-Based Wearable Sensor for Human Motion Intent Recognition

• Published in: IEEE/ASME transactions on mechatronics Vol. 29; no. 6; pp. 4120 - 4130
• Main Authors: Zheng, Enhao, Wan, Jiacheng, Hu, Nanxing, Wang, Qining
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.12.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Fabric-based sensors; Fabrication; Feature extraction; Force; Grip force; Human activity recognition; Human motion; human motion recognition; Limbs; Motion perception; muscle features; Muscles; Robot control; Robot dynamics; Robot sensing systems; Sensors; Shape control; Time response; wearable sensing; Wearable sensors; Wearable technology
• ISSN: 1083-4435; 1941-014X
• DOI: 10.1109/TMECH.2024.3363454

Fabric-based wearable sensing is receiving increasing attention in the field of wearable robots. In our study, we propose a fabric-based sensing method for human motion recognition/estimation. The approach was developed with an elastic sleeve integrated with four bend sensors and the superellipse-based construction algorithm. Unlike existing techniques, our method can extract muscular geometrical features in the anatomical cross-sectional plane. To validate our method, we conducted evaluations on 14 subjects, including time response evaluations, isometric grip force estimation, forearm/lower limb joint angle estimation, discrete lower limb posture recognition, and continuous gait phase estimation. First, our method produced comparable results to the state-of-the-art approaches. The average <inline-formula><tex-math notation="LaTeX">R^{2}</tex-math></inline-formula> values for joint angle estimation were 0.84-0.94, the average accuracy for lower limb posture recognition was 99.78%, and the average estimation error for gait phase was below 1% of a complete gait cycle. Second, we accomplished tasks that existing fabric-based mechanical sensors are unable to achieve. We demonstrated that our method detected motion onsets before the actual joint movements in voluntary dorsiflexion and sit-to-stand transition tasks. In addition, we achieved isometric grip force estimation with an average <inline-formula><tex-math notation="LaTeX">R^{2}</tex-math></inline-formula> of 0.89. Unlike stretch-based methods that measure the response of movements, our method extracts human motion intents before the actual movements occur. This extends the measurement scope of fabric-based wearable sensing for human motion recognition. In future work, we will focus on sensor integration and robot control to further enhance our method's capabilities.