Human Activity Recognition from Triaxial Accelerometer Data

The demand for objectivity in clinical diagnosis has been one of the greatest challenges in Biomedical Engineering. The study, development and implementation of solutions that may serve as ground truth in physical activity recognition and in medical diagnosis of chronic motor diseases is ever more imperative. This paper describes a human activity recognition framework based on feature extraction and feature selection techniques where a set of time, statistical and frequency domain features taken from 3-dimensional accelerometer sensors are extracted. In this paper, unsupervised learning is applied to the feature representation of accelerometer data to discover the activities performed by different subjects. A feature selection framework is developed in order to improve the clustering accuracy and reduce computational costs. The features which best distinguish a particular set of activities are selected from a 180th- dimensional feature vector through machine learning algorithms. The implemented framework achieved very encouraging results in human activity recognition: an average person-dependent Adjusted Rand Index (ARI) of 99:29%-0:5% and a person-independent ARI of 88:57%-4:0% were reached.