Comparison of Wearable Sensor Based Algorithms for Upper Limb Activity Detection

Upper limb activity detection using wearable sensors is useful for continuous monitoring in rehabilitation. In this study, we analysed four popular algorithms that compute real world amount of arm use using wrist triaxial accelerometry or inertial measurement units, and compared them to “actual” arm-use identified from videos by two independent assessors. It was found that the accelerometry-based methods are sensitive to arm movements, but are poor at distinguishing functional and non-functional movements. Use of arm orientation information makes the arm-use estimation robust to overestimation due to several non-functional movements, while losing some sensitivity to functional movements. A merger of principles from these two methods might result in a more accurate approach for arm-use detection.