Machine-Learning-Based Human Motion Recognition via Wearable Plastic-Fiber Sensing System
Shuang Wang, Bin Liu, Yu-Lin Wang, Yingying Hu, Juan Liu, Xingdao He, Jinhui Yuan, Qiang Wu
Abstract
Wearable human–machine interface (HMI) is a medium for information transmission and exchange between people and computers. It is widely used in the fields of human motion capture and recognition and augmented/virtual reality (AR/VR). This research proposes a wearable plastic-optical-fiber (POF) sensing system based on machine learning for human motion recognition. The wearable sports sleeve is designed and worn on the elbow and knee joints of human body. The wearable sensor system uses a D-shaped POF (DPOF) sensor, whose coefficient of determination (R 2) is 0.96496 and sensitivity is -0.7859% per degree. Support vector machines (SVMs), MobileNetV2 network, and transfer learning were used to identify six types of movement: walking, running, going upstairs, going downstairs, high leg lifts, and rope skipping. The accuracy of classification based on the four joint position monitoring can reach 98.28%, 98.94%, and 99.74%, respectively. The proposed POF wearable system has good applications for human motion state recognition and possesses great application potential in AR/VR.