運用機器學習演算法於肌電圖手勢訊號辨識

鄭宜崴 ； 何文獻 ； 鄭雅倫 ； 陳以德

《NCS 2017 全國計算機會議》 2017 (2018/01) Pp. 909-912

https://doi.org/10.29428/9789860544169.201801.0170

摘要

近年來，新穎的人機介面如雨後春筍般湧現，如Wii的搖桿以慣性感測器為基礎，Kinect、Xtion, Leap Motion等基於影像視覺的技術，或是基於聲音的語音辨識…等等非常多樣化。感測器中Myo Armband為加拿大新創公司Thalmic Labs 開發的一款智能手環，內建九軸慣性感測器(IMU)、八組肌電感測器(EMG)，使用低功耗藍芽進行通訊，Myo內建手勢分類器可分類出手掌之「放鬆」、「握拳」、「張開」、「內彎」、「外彎」、「雙擊」等姿勢。然而，Myo作為一項創新的發明，其分類演算法仍未盡善盡美，時常發生手勢辨別錯誤的問題，如「握拳」和「張開」就經常發生誤判;另外，Myo可辨別之手勢僅有上述之六種，作為一項人機介面則顯得略少。因此，本研究旨在應用機器學習的演算法，建立一套手勢分類器，期望能改善Myo手環之手勢辨識準確率及增加可識別之手勢種類。

關鍵字

人機介面 ； Myo手環 ； 肌電圖 ； 機器學習 ； 手勢辨識

並列摘要

In recent years, the new Human-Computer Interface has sprung up, such as the inertial sensor-based Wii Joystick, the image-based visual technology Kinect, Xtion, Leap Motion and voice-based voice recognition ... and so on. Myo is a smart armband developed by Thalmic Labs in Canada. Myo built-in 9-axis inertial sensor, 8 electromyography (EMG) sensors, and communicate with low-power Bluetooth. Analyzing the surface EMG signal (sEMG) can recognize the user's hand gesture. And MYO built-in gesture classifier can be classified as the palm of the 〞relax〞, 〞fist〞, 〞open〞, 〞bend〞, 〞outside the bend〞, 〞double hit〞. However, as an innovative invention, Myo’s classification algorithm is still not perfect. Wrong gesture recognition often happened; such as 〞Fist〞 and 〞open〞 are often misjudged. In addition, Myo can recognize only six gestures, as a human-computer interface is slightly less. Therefore, this study aims to use the machine-learning algorithm to build a set of gesture classifier, hoping to improve the Myo Armband’s gesture recognition accuracy and increase the number of recognizable gestures.

並列關鍵字

Human-Computer Interface ； Myo Armband ； EMG ； Machine Learning ； Hand Gesture Recognition