Quentin Mascret1, Mathieu Bielmann2, Fall Cheikh Latyr, Laurent Bouyer1, Benoit Gosselin3
13:30 - 15:00 | Tue 30 Oct | Ambassador C | B4L-A.5
In the realm of Human Activity Recognition (HAR),supervised machine learning and deep learning are commonlyused. Their training is done using time and frequency featuresextracted from raw data (inertial and gyroscopic). Nevertheless,raw data are seldom employed. In this paper, a dataset ofable-bodied participants is recorded using 3 custom wirelessmotion sensors providing embedded IMU and sEMG detectionand processing and a base station (a Raspberry Pi 3) running aclassification algorithm. A Support Vector Machine with RadiusBasis Function Kernel (RBF-SVM) is augmented using SphericalNormalization to achieve a motion classification accuracy of97.68% between 8 body motions. The proposed classifier allowsfor real-time prediction callback with low latency output.