Classification of Daily Human Activities Using Wearable Inertial Sensor

Abstract

Classification of daily human activities using wearable inertial sensors is presented. Two sensing devices namely the accelerometer sensor mounted on arduino controller and shimmer device are used for acquiring data. Data are acquired from thirty eight healthy subjects without any form of disabilities. Variation in classification accuracy considering data obtained from shimmer device, accelerometer sensor and combination of shimmer & accelerometer data are analysed. Performance of two classifiers namely the KNN classifier and SVM classifier in classifying actions are tested. Various experimental analyses proves that among the data considered for classification, combination of shimmer data and accelerometer data provided better results. Also KNN classifier is found to perform better with an average overall accuracy of 95.6% which is around 6% higher that the accuracy obtained with SVM classifier.