Recognizing Manual Activities Using Wearable Inertial Measurement Units: Clinical Application for Outcome Measurement

Abstract

The ability to monitor activities of daily living in the natural environments of patients could become a valuable tool for various clinical applications. In this paper, we show that a simple algorithm is capable of classifying manual activities of daily living (ADL) into categories using data from wrist- and finger-worn sensors. Six participants without pathology of the upper limb performed 14 ADL. Gyroscope signals were used to analyze the angular velocity pattern for each activity. The elaboration of the algorithm was based on the examination of the activity at the different levels (hand, fingers and wrist) and the relationship between them for the duration of the activity. A leave-one-out cross-validation was used to validate our algorithm. The algorithm allowed the classification of manual activities into five different categories through three consecutive steps, based on hands ratio (i.e., activity of one or both hands) and fingers-to-wrist ratio (i.e., finger movement independently of the wrist). On average, the algorithm made the correct classification in 87.4% of cases. The proposed algorithm has a high overall accuracy, yet its computational complexity is very low as it involves only averages and ratios.

Keywords

• manual activities
• activity recognition
• movement classification
• remote health monitoring
• outcome assessment
• questionnaires