Title: Human activity detection from inertial data using RNN and LSTM network

Authors: Thanina Boultache; Brahim Achour; Mourad Laghrouche

Addresses: Laboratoire d'Analyse et Modélisation des Phénomènes Aléatoires, Mouloud Mammeri University, BP 17 RP, 15000, Tizi-Ouzou, Algeria ' Laboratoire de Recherche en Informatique, Mouloud Mammeri University, BP 17 RP, 15000, Tizi-Ouzou, Algeria ' Laboratoire d'Analyse et Modélisation des Phénomènes Aléatoires, Mouloud Mammeri University, BP 17 RP, 15000, Tizi-Ouzou, Algeria

Abstract: With the advancement of smartphones and the rapid development of artificial intelligence, human activity detection systems help to improve human welfare and health. However, these systems need to be constantly renewed, improved and updated. In this paper, we propose an automatic and real-time classification system of physical activities using deep learning architecture, recurrent neural networks (RNN) and short-term memory networks (LSTM). To develop and validate the learning model we gathered a dataset of 697,964 recordings of accelerometer and gyroscope signals embedded in smartphones. The data was collected from 20 people aged 13 to 82 years old of different categories (sportsman, elderly, etc.). Then, the raw data is preprocessed and the learning model is trained. The results of the experiment show that the proposed algorithm achieves an average accuracy of 95%. The developed model was exported to an Android application for real-time predictions.

Keywords: artificial intelligence; deep learning; recurrent neural networks; RNN; long-short-term memory; LSTM; activity recognition; smartphones; accelerometer.

DOI: 10.1504/IJSNET.2022.124568

International Journal of Sensor Networks, 2022 Vol.39 No.3, pp.156 - 161

Received: 12 Dec 2021
Accepted: 04 Jan 2022
Published online: 28 Jul 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article