Scholarlink

<p>In the field of WiFi Sensing (WS), developing applications requires data with quality, quantity, and variability to enhance cross-domain capability.This paper presents EHUNAM, a comprehensive channel state information (CSI) dataset developed for various WS applications, with a primary focus on people counting (PC), human activity recognition (HAR), and machine activity recognition (MAR), while remaining suitable for additional tasks. The dataset was acquired using diverse equipment configurations and under different scenarios, ensuring versatility and representativeness. Beyond traditional applications, EHUNAM includes measurements for recognizing activities of home appliances and industrial machines. To achieve high accuracy in new settings, data was collected over 23 days in eight distinct environments, including an industrial scenario, involving 21 people and nine machines that can also perform activities simultaneously. Validation using a convolutional neural network (CNN) for PC, HAR, and machine activity recognition (MAR), considering multiclass and multilabel classification, achieved over 90% accuracy in most cases, underscoring the dataset’s robustness and its capacity to tackle a broad spectrum of real-world scenarios.</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

EHUNAM, a WiFi CSI-based dataset for human and machine sensing

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月17日

查看全文
  • Elizabet de Armas,
  • Guillermo Diaz,
  • Iker Sobron,
  • Iñaki Eizmendi,
  • Iratxe Landa,
  • Jose Maria Matias,
  • Manuel Velez

摘要

In the field of WiFi Sensing (WS), developing applications requires data with quality, quantity, and variability to enhance cross-domain capability.This paper presents EHUNAM, a comprehensive channel state information (CSI) dataset developed for various WS applications, with a primary focus on people counting (PC), human activity recognition (HAR), and machine activity recognition (MAR), while remaining suitable for additional tasks. The dataset was acquired using diverse equipment configurations and under different scenarios, ensuring versatility and representativeness. Beyond traditional applications, EHUNAM includes measurements for recognizing activities of home appliances and industrial machines. To achieve high accuracy in new settings, data was collected over 23 days in eight distinct environments, including an industrial scenario, involving 21 people and nine machines that can also perform activities simultaneously. Validation using a convolutional neural network (CNN) for PC, HAR, and machine activity recognition (MAR), considering multiclass and multilabel classification, achieved over 90% accuracy in most cases, underscoring the dataset’s robustness and its capacity to tackle a broad spectrum of real-world scenarios.