손짓이 명령어가 된다…부산대, 손동작 10가지 94.5% 정확도로 읽는 웨어러블 개발

Translation

Pusan National University announced on 21 September that the research teams of Professors 심현석 and 박성윤 in its School of Electrical and Electronics Engineering had jointly developed a wearable bio-interface system that combines stretchable bio-electrodes with a tiny, low-power semiconductor chip. Using electrodes attached to the arm, the team measured the electromyographic signals that change with hand movements and analysed them with machine learning. As a result, it classified ten hand gestures with an average accuracy of 94.5%. The electrodes, made from a conductive polymer, keep their electrical properties even when stretched by 40% and performed stably through 1,000 cycles of repeated stretching. A 20-channel electrode array was combined with a 32-channel low-power application-specific integrated circuit (ASIC), 'NR32A1', which integrates amplification, filtering and conversion of bio-signals into digital form and is 47% smaller in area and uses 30% less power than existing commercial bio-signal chips. The team built a real-time calculator control in which gestures stand for digits and arithmetic signs, so that sums can be entered and computed by hand movements alone, and a function that turns particular gestures into expressions such as 'hungry' and 'thirsty'; the results were published in the international journal 'Small' on 28 August.