Orientation-based classification of sit-to-stand activity using Artificial Neural Network in real-time
1 Human-Machine Interaction Research Center, Informatics Engineering, Faculty of Science and Technology, Ma Chung University, Jl. Villa Puncak Tidar N-01, Malang 65151, Indonesia.
2 Informatics Engineering, Faculty of Industrial Technology, National Institute of Technology Malang, Jl. Raya Karanglo km.2, Malang, Indonesia.
3 Graduate School of Life Science and Systems Engineering, Kyushu Institute of Technology, 2-4 Hibikino- Wakamatsu-ku- Kitakyushu, Fukuoka 808-0196, Japan.
Research Article
World Journal of Advanced Research and Reviews, 2023, 17(03), 307–317
Article DOI: 10.30574/wjarr.2023.17.3.0376
Publication history:
Received on 31 January 2023; revised on 03 March 2023; accepted on 06 March 2022
Abstract:
Human activity recognition is highly required to develop an assistive technology. This study proposes the use of Artificial Neural Network (ANN) to classify Sit to Stand (STS) activity based on the sensor's orientation angle. There are four main phases in this research which are Sit Phase, Flexion Phase, Extension Phase, and Stabilization Phase (Stand). Human activity recognition is highly required to develop an assistive technology. STS activity is an important movement for every human being despite the inability of certain age groups to perform this movement due to weakened muscle function. The limited information from previous on the difficult phases experienced by the subjects to perform STS causes the development process of assistive devices slower. Our solution can classify those phases in real-time using the angles on Korpus Sterni (chest) and Tibia (calf) to gather information on which phase is difficult to be performed. It manages to gather and process the sensor data on application with approximately 3 seconds delay, resulting in the extension phase being a difficult phase to classify. A dataset of 32,000 samples was obtained from 8 subjects consisting of 6 subjects aged 20-30 years and 2 subjects aged 40-50 years. After experimenting and testing the performance of the ANN architecture, the neural network architecture consisted of 4 input nodes, 4 hidden layers (93-69-89-76) with appropriate hyperparameters, and 4 output layers. The training accuracy and testing accuracy reached 86% and 72% respectively.
Keywords:
ANN; HAR; IMU Sensor; STS activity
Full text article in PDF:
Copyright information:
Copyright © 2023 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution Liscense 4.0