Scholarlink

<p>Fall detection is crucial for monitoring elderly safety, and radar-based non-contact sensing has shown significant promise in this field. However, the literature lacks comprehensive analyses on optimizing radar sensor positioning and utilizing shorter radar segments for improved accuracy. In this work, we propose a methodology for multidomain multi-radar-based fall detection using 1D and 2D convolutional neural networks (CNN). For this, an online publicly available radar database (N=15) is considered. The radar signals are segmented, and range time (RT), range Doppler (RD), and Doppler time (DT) maps are obtained. These maps were fed to 1D and 2D CNN, and features were fused and given to the classifier. The performance is evaluated using 10-fold cross-validation. The proposed approach is able to discriminate elderly fall. The multidomain fusion model yields average classification accuracy and F1-score of 85.95% and 86%, respectively. Thus, the proposed approach could be extended for ubiquitous, real-time in-home monitoring and fall detection.</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

Automated radar-based elderly fall detection using novel multidomain fusion approach

  • 期刊论文

  • 发表日期：

    2025年11月17日

查看全文
  • P. J. Swarubini,
  • Tomohiko Igasaki,
  • P. A. Karthick,
  • Nagarajan Ganapathy

摘要

Fall detection is crucial for monitoring elderly safety, and radar-based non-contact sensing has shown significant promise in this field. However, the literature lacks comprehensive analyses on optimizing radar sensor positioning and utilizing shorter radar segments for improved accuracy. In this work, we propose a methodology for multidomain multi-radar-based fall detection using 1D and 2D convolutional neural networks (CNN). For this, an online publicly available radar database (N=15) is considered. The radar signals are segmented, and range time (RT), range Doppler (RD), and Doppler time (DT) maps are obtained. These maps were fed to 1D and 2D CNN, and features were fused and given to the classifier. The performance is evaluated using 10-fold cross-validation. The proposed approach is able to discriminate elderly fall. The multidomain fusion model yields average classification accuracy and F1-score of 85.95% and 86%, respectively. Thus, the proposed approach could be extended for ubiquitous, real-time in-home monitoring and fall detection.