In this project we propose the MyoTrack system, which uses the Myo surface electromyography device to augment arm tracking accuracy rates achieved by IMU strategies, while maintaining the privacy and convenience of using mobile devices. We present a classification routine that can track different arm poses with a mean accuracy of 93.79\%. We demonstrate reasonable accuracy in recognizing the EMG signatures related to maintaining the arm stationary in various postures/regions, which can in turn be used to enhance IMU-based strategies.
We define a set of poses, that when combined, would serve as an estimate of the user's arm location. It should be noted that poses described in this study do not require the user to explicitly perform a descriptive gesture. Rather, the system identifies these poses when it observes that the arm is stationary and uses them for arm localization. MyoTrack uses two sets of recognition schemes to obtain the said localization; viz. - region estimation and elbow angle estimation. Region estimation refers to a coarse estimation that seeks to identify the region of operation of the user's arm. Once a coarse estimation of the region is obtained, we move on to the elbow angle estimation module of MyoTrack.
The elbow angle estimation routine comprises of 4 different poses described in figure below. These poses refer to the fine estimation of the user's arm position. The system seeks to identify landmark poses that can be used as a means to re-calibrate the gyroscope when identified. The elbow estimation routine essentially classifies any given pose into either landmark locations (Fig - D) or non-landmark location (Fig - A,B,C). We hypothesize that the aforementioned region and elbow angle estimation techniques along with the state of art IMU tracking systems can serve as a reliable means for 3D arm tracking.