Abstract
In this paper, we describe and evaluate an activity recognition system using a single 3-axis accelerometer and a barometric sensor worn on a waist of the body. The purpose of this work is to prevent child accidents such as unintentional injuries at home. In order to prevent child accidents in the home and reduce efforts of parents, we present a new safety management system for babies and children. We collected labeled accelerometer data from babies as they performed daily activities which are standing still, standing up, sitting down, walking, toddling, crawling, climbing up, climbing down, stopping, wiggling, and rolling. In order to recognize daily activities, mean, standard deviation, and slope of time-domain features are calculated over sliding windows. In addition, the FFT analysis is adopted to extract frequency-domain features of the aggregated data, and then energy and correlation of acceleration data are calculated. We used the resulting training data to induce a predictive model for activity recognition. Naive Bayes, Bayes Net, Support Vector Machine, k-Nearest Neighbor, Decision Tree, Decision Table, Multilayer Perceptron, Logistic classifiers are tested on these features. Classification results using training and eight classifiers were compared. The overall accuracy of activity recognition was 96.2% using only a single wearable triaxial accelerometer sensor with the k-Nearest Neighbor.