Hee-Tae Jung1, Park Joon Woo, Jeong Jugyeong Jeong Jugyeong2, Taekyeong Ryu2, Yangsoo Kim2, Sunghoon Ivan Lee3
10:20 - 10:35 | Mon 5 Mar | Treasure Island ABC | MoAT1.4
When stroke survivors perform rehabilitation exercises in clinical settings, experienced therapists can evaluate the associated quality of movements by observing only the initial part of the movement execution so that they can discourage therapeutically undesirable movements effectively and reinforce desirable ones as much as possible in the limited therapy time. This paper introduces a novel monitoring platform based on wearable technologies that replicates such capability of skilled therapists. Specifically, we propose to deploy five wearable sensors on the trunk, and upper and forearm of the two upper limbs, analyze partial to complete observation data of reaching exercise movements, and employ supervised machine learning to estimate therapists' evaluation of movement quality. Estimation performance was evaluated using F-Measure, Receiver Operating Characteristic Area, and Root Mean Square Error, showing that the proposed system can be trained to evaluate the movement quality of the entire exercise using as small as the initial $5s$ of the exercise performance. The proposed platform may help ensure high quality exercise performance and provide virtual feedback of experienced therapists during at-home rehabilitation.