| Title |
Transfer Learning-based Human Motion Capture and Classification System using Multiple Low-cost IMU and Flex Pressure Sensors |
| Authors |
배문규(Munkyu Bae) ; 김건태(Guntae Kim) ; 박이형(Yi-Hyeong Park) ; 강창묵(Chang Mook Kang) |
| DOI |
https://doi.org/10.5370/KIEE.2023.72.11.1434 |
| Keywords |
Healthcare; Wearable device; IMU; Motion classification; Transfer learning |
| Abstract |
In this paper, we present a transfer learning-based human motion capture and classification system using multiple IMU and flex sensors without camera. The proposed system consists of 6 low-cost IMUs and 12 flex pressure sensors. For the human motion classification, the signals of multiple IMUs were reconstructed into images using wavelet transform. The transformed image is designed to be used as an input to the convolution neural network. The network was designed using Inception V3-based transfer learning. For motion capture without using a camera, the presented system was constructed through a combination of IMUs (MPU6050) and flex pressure sensors (thin film pressure sensors) rather than multiple IMUs. The proposed method showed 86.67% accuracy and 0.869 points of f1-score. The proposed system is expected to be applicable to a healthcare system that can provide exercise coaching without cameras, and without space restrictions. |