Identificação de atividades cotidianas com membros superiores através de sensores inerciais

Abstract

In this work, the development of data processing strategies and methodologies for identifying 7 everyday activities performed with the upper limbs was proposed, with the support of inertial sensors and machine learning techniques. Computational methods were used for data processing, including data synthesis to homogenize the number of samples, digital filter bank, Hamming windowing, feature extraction, and classification using a neural network with the Matlab Neural Net Pattern Recognition toolbox. In this study, two units of inertial measurement were used, one on each wrist. This choice was made because a significant portion of the performed activities requires cooperation between the two limbs. Thus, through supervised training of the neural networks, an average accuracy of over 90% was achieved for activity identification, demonstrating the feasibility of the proposed strategies for identifying daily activities involving the upper limbs.