Human Activities Recognition Using Accelerometer and Gyroscope

Plenty of supervised machine learning techniques that use accelerometer and gyroscope signals for automatic Human Activity Recognition (HAR) has been proposed in the last decade. According to recent studies, the combination of accelerometer and gyroscope signals, also called multimodal recognition, increases the accuracy in HAR with respect to the use of each signal alone. This paper presents the results of an analysis we performed in order to compare the effectiveness of machine learning techniques when used separately or jointly on accelerometer and gyroscope signals. We compare SVM and \(k-\)NN classifiers (combined with hand-crafted features) with a deep residual network using three publicly available datasets. The results show that the use of deep learning techniques in multimodal mode (i.e., using accelerometer and gyroscope signals jointly) outperforms other strategies of at least 10%.