Scholarlink

<p>Human activity recognition (HAR) typically involves detecting and deducting activities such as walking, running, and lying down. Nevertheless, recognizing these activities is insufficient to provide comprehensive information about the user. Inferring the behavioral context associated with these activities is crucial for a better understanding of the environment of the user. To this end, multiple Machine Learning (ML) models are often used: one model to recognize activities and the other to identify the behavioral context. However, deploying multiple models leads quickly to significant resource consumption, including execution time, memory, and energy. To address this challenge, we propose a system capable of recognizing different activities and their associated behavioral contexts using a single ML model based on extremely randomized trees to reduce resource consumption. It uses inertial data and contextual information from everyday smartphones. Evaluation based on the HAR ExtraSensory public dataset demonstrated the effectiveness of the proposed approach, achieving 97.10% accuracy while significantly reducing resource consumption. More specifically, the system achieved over a 97% reduction in execution time, a 78% reduction in energy consumption, and a near reduction of 72% in memory usage compared to a competing method in the literature. The code is publicly available in: <a href="https://github.com/hocine12300/Smartphone-Sensor-Fusion-for-Joint-Recognition-of-Human-Activity-and-Behavioral-Context-">https://github.com/hocine12300/Smartphone-Sensor-Fusion-for-Joint-Recognition-of-Human-Activity-and-Behavioral-Context-</a>.</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

Smartphone sensor fusion for joint recognition of human activity and behavioral context

  • 期刊论文

  • 发表日期：

    2025年9月10日

查看全文
  • Hocine Attoumi,
  • Achour Achroufene,
  • Redouane Saifi,
  • Lydia Souici,
  • Djamila Boukredera

摘要

Human activity recognition (HAR) typically involves detecting and deducting activities such as walking, running, and lying down. Nevertheless, recognizing these activities is insufficient to provide comprehensive information about the user. Inferring the behavioral context associated with these activities is crucial for a better understanding of the environment of the user. To this end, multiple Machine Learning (ML) models are often used: one model to recognize activities and the other to identify the behavioral context. However, deploying multiple models leads quickly to significant resource consumption, including execution time, memory, and energy. To address this challenge, we propose a system capable of recognizing different activities and their associated behavioral contexts using a single ML model based on extremely randomized trees to reduce resource consumption. It uses inertial data and contextual information from everyday smartphones. Evaluation based on the HAR ExtraSensory public dataset demonstrated the effectiveness of the proposed approach, achieving 97.10% accuracy while significantly reducing resource consumption. More specifically, the system achieved over a 97% reduction in execution time, a 78% reduction in energy consumption, and a near reduction of 72% in memory usage compared to a competing method in the literature. The code is publicly available in: https://github.com/hocine12300/Smartphone-Sensor-Fusion-for-Joint-Recognition-of-Human-Activity-and-Behavioral-Context-.