Image-based Activity Recognition from IMU Data

Abstract

In this paper we propose an approach to improve the performance of human activity recognition (HAR) from inertial sensors data, based on image processing techniques. Our approach creates image representations of the time-series data to take advantage of the strengths that convolutional neural networks (CNNs) have shown when dealing with image data. We have conducted an evaluation using benchmark datasets that are considered among the most relevant in HAR. Our results show that our approach is able to outperform the state of the art in all cases.