電気インピーダンストモグラフィによる前腕筋骨格運動情報を用いた手指姿勢推定

抄録

For human-machine cooperation and interfaces, the need to accurately estimate hand and finger postures and movements without restricting action is increasing. Clarifying the relationship between hand and finger postures and musculoskeletal activity will help understand human motor function and biological mechanism modeling. This study aimed to develop a system to estimate hand-finger postures from the tissue distribution of a wrist section obtained by using electrical impedance tomography (EIT) and reveal the influence of the musculoskeletal features on estimation performance. Specifically, this study proposed a method to classify hand postures using a Support Vector Machine (SVM) based on the reconstructed relative conductivity distribution of a wrist cross-section using EIT, and compared its accuracy with the conventional classification methods using original voltage data, i.e., EIT signals. In the experiment, a band with 16 electrodes was placed around the wrists of 6 participants and EIT signals were obtained for 7 different hand postures. The reconstructed musculoskeletal activity-based classification performed better than EIT voltage signal-based classification, achieving the highest accuracy of 96.1%. The spatial distribution of images from the reconstruction of musculoskeletal activity also reflected the positions of the flexors and extensors, with different activities observed for each posture.