An Online Human Activity Recognizer for Mobile Phones with Accelerometer

摘要

We propose a novel human activity recognizer for an application for mobile phones. Since such applications should not consume too much electric power, our method should have not only high accuracy but also low electric power consumption by using just a single three-axis accelerometer. In feature extraction with the wavelet transform, we employ the Haar mother wavelet that allows low computational complexity. In addition, we reduce dimensions of features by using the singular value decomposition. In spite of the complexity reduction, we discriminate a user's status into walking, running, standing still and being in a moving train with an accuracy of over 90%.