Scholarlink

<p>Automatic activity recognition systems aim to acquire the user’s state and environment using mixed sensors and allow continuous monitoring of various physiological signs, with these sensors connected to the subject’s body. This is highly beneficial for healthcare applications, including intelligent, automated daily activity monitoring for individuals with disabilities and the elderly. Human activity recognition (HAR) is a dynamic research field for classifying human movement and its applications across various areas, including medical diagnosis, healthcare systems, smart home monitoring, and elderly care. HAR data are collected from wearable devices that include numerous types of sensors or mobile sensor support. Now, deep learning (DL) is primarily used to develop HAR methods that discover human activity patterns from sensor data to help people with disabilities. In this paper, a Human Activity Recognition Model for Disabilities via an Osprey Optimisation Algorithm with Dimensionality Reduction (HARD-OOADR) model is proposed. The aim is to investigate the feasibility of developing an assistive tool for automatic daily activity monitoring of people with disabilities. At first, the data pre-processing stage uses the linear scaling normalisation (LSN) method to prepare and refine raw data by cleaning, transforming, and organising it for analysis. Furthermore, the HARD-OOADR model employs the equilibrium optimisation (EO) method for feature selection. For classification, the bidirectional gated recurrent unit with attention (BiGRU-A) model is used. To optimise the BiGRU-A classifier’s parameters, the osprey optimisation algorithm (OOA) is used. A comprehensive experimental analysis of the HARD-OOADR model is performed using a smartphone dataset. The comparison study of the HARD-OOADR model showed an accuracy of 99.18% compared to existing approaches.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive motion assisted human activity recognition for people with disabilities via osprey optimisation-based dimensionality reduction with recurrent neural network

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Mofadal Alymani,
  • Munya A. Arasi,
  • Hussain Alshahrani,
  • Amani A. Alneil

摘要

Automatic activity recognition systems aim to acquire the user’s state and environment using mixed sensors and allow continuous monitoring of various physiological signs, with these sensors connected to the subject’s body. This is highly beneficial for healthcare applications, including intelligent, automated daily activity monitoring for individuals with disabilities and the elderly. Human activity recognition (HAR) is a dynamic research field for classifying human movement and its applications across various areas, including medical diagnosis, healthcare systems, smart home monitoring, and elderly care. HAR data are collected from wearable devices that include numerous types of sensors or mobile sensor support. Now, deep learning (DL) is primarily used to develop HAR methods that discover human activity patterns from sensor data to help people with disabilities. In this paper, a Human Activity Recognition Model for Disabilities via an Osprey Optimisation Algorithm with Dimensionality Reduction (HARD-OOADR) model is proposed. The aim is to investigate the feasibility of developing an assistive tool for automatic daily activity monitoring of people with disabilities. At first, the data pre-processing stage uses the linear scaling normalisation (LSN) method to prepare and refine raw data by cleaning, transforming, and organising it for analysis. Furthermore, the HARD-OOADR model employs the equilibrium optimisation (EO) method for feature selection. For classification, the bidirectional gated recurrent unit with attention (BiGRU-A) model is used. To optimise the BiGRU-A classifier’s parameters, the osprey optimisation algorithm (OOA) is used. A comprehensive experimental analysis of the HARD-OOADR model is performed using a smartphone dataset. The comparison study of the HARD-OOADR model showed an accuracy of 99.18% compared to existing approaches.