End-to-End Human Activity Recognition using Convolutional Neural Network and Center Loss

Abstract

© 2020, World Academy of Research in Science and Engineering. All rights reserved. Sensor-based human activity recognition (HAR) is an interesting research direction in the fields of healthcare, virtual reality, and other domains. In recent years, deep learning has been grown rapidly, and they have many successes in a wide range of domains such as computer vision, intelligent transportation, and human activity recognition. Previous studies that used deep learning-based methods to tackle the problem of human activity recognition did not consider the embedded features extracted from deep learning architectures, so it is hard to classify activities with similar patterns. In this paper, we would like to propose a deep learning-based method that takes the merits of Convolutional Neural Network (CNN) and Center Loss to recognize daily living activities. By stacking multiple CNN layers, we obtained an architecture robust for extracting features from sensor data, and applying Center Loss on embedded features makes our method more robust to discriminate between classes that have similar patterns. In experimental results, the proposed method achieves the accuracy rate of 94.2% F1-Score on the smartphone dataset.