Monitoring and machine learning-based classification of human activities in bed using wireless devices and a fusion of acceleration and sEMG signals

This research aims to develop a system for monitoring and classifying human activities in bed using three-axis accelerometer (ACM) and surface electromyography (sEMG) signals. The contributions of this work are that, first, we develop and implement 2.4 GHz IEEE 802.15.4 wireless sensor nodes combined with a three-axis ACM (i.e., GY-521) and an sEMG sensor (i.e., OYMotion), where sensor data are wirelessly sent to a receiver connected to a computer for processing. Second, nine human activities in bed, including rapid breathing, seizure sleeping, and falling from the bed, as the critical events, are considered. Third, human activity classification is carried out using a machine learning-based classification framework with 150 features and six classifiers with several sub-functions, including Decision Tree (DT), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Naive Bayes (NB), Neural Networks (NN), and Ensemble. Three cases of input data for classification are evaluated: only ACM data (for motion), only sEMG data (for muscle contraction), and fusion data. Experimental results demonstrate that three-axis ACM and sEMG data are successfully sent via wireless communications for both line-of-sight (LOS) and non-line-of-sight (NLOS) environments, where efficient monitoring can be achieved. Additionally, we can obtain 98 % classification accuracy when both sensor data and the Ensemble Subspace KNN method are used. Specifically, we can accurately detect abnormal events such as rapid breathing, seizure sleeping, falling from the bed, and lying down on the ground, with an accuracy of 98.8 %, 97.7 %, 92.0 %, and 99.3 %, respectively.