Elderly Activity Recognition Based Embedded Machine Learning

This paper presents the implementation of an embedded-based machine learning system for elderly activity recognition. Two types of neural networks namely Convolutional Neural Network (CNN) and Multilayer Perceptron (MLP) were used to process data from Inertial Measurement Units (IMUs) sensors mounted on the legs of elderly participants. Results demonstrate the effectiveness of the proposed system in classifying elderly activity, achieving an accuracy of 99% and 80% for CNN and MLP, respectively. The MLP performs better in inference time than the CNN with 2 ms per classification compared to 656 ms for CNN. The proposed system has been successfully deployed on an Arduino nano BLE sense providing real-time on-board elderly activity recognition.