Research on Activity Recognition of People with Mobility Disabilities using Personal Smart Devices

Abstract

Human activity recognition (HAR) classifies and interprets people's movements using sensors on wearable smart devices, offering potential benefits to diverse populations such as general public, elderly, and patients. However, there is limited research on the activities of people with mobility disabilities. In this study, we collected data from 60 participants, analyzing fundamental activities like still, walking, crutches, walker, manual wheelchair, and electric wheelchair using smartphone and smartwatch sensors. Our analysis focused on recognition tasks, sensor combinations, classification models, evaluation methods, and integrating smartphone and smartwatch data. Results showed high classification accuracies of 99.34% using the random evaluation method and 97.94% in the user-independent evaluation method with optimal sensor combinations. This study contributes to expanding HAR research for individuals with mobility disabilities and fostering interest in related applications.