Wi-Fi and Radar Fusion for Head Movement Sensing Through Walls Leveraging Deep Learning
Wi-Fi and Radar Fusion for Head Movement Sensing Through Walls Leveraging Deep Learning
复制标题
DOI：
10.1109/jsen.2023.3337515
复制
发表时间：
2024-05
影响因子：
4.3
通讯作者：
Hira Hameed;Ahsen Tahir;Muhammad Usman;Jiang Zhu;.. Lubna-Lubna-2213347557;H. Abbas;N. Ramzan;T. Cui;M. Imran;Q. Abbasi
中科院分区：
文献类型：
--
作者：
Hira Hameed;Ahsen Tahir;Muhammad Usman;Jiang Zhu;.. Lubna-Lubna-2213347557;H. Abbas;N. Ramzan;T. Cui;M. Imran;Q. Abbasi
头部运动的检测在人机交互系统中起着至关重要的作用。这些系统依赖于控制信号来操作一系列辅助和增强技术，包括用于四肢瘫痪者的轮椅，以及虚拟/增强现实和辅助驾驶。驾驶员困倦检测和头部运动检测辅助的警报系统可以防止重大事故并挽救生命。可穿戴设备，如MagTrack，由磁性标签和磁性眼镜夹组成，具有侵入性。基于视觉的系统受到环境照明、视线和隐私问题的影响。非接触式传感已成为下一代传感和检测技术的重要组成部分。Wi-Fi和雷达提供非接触式传感，但是，在辅助驾驶中，它们需要在外壳或仪表板内，出于本文中的所有实际目的，它们被认为是通过墙壁。在这项研究中，我们提出了一个非接触式系统来检测人体头部运动与墙壁。我们使用超宽带（UWB）雷达和Wi-Fi信号，利用机器和深度学习（DL）技术。我们的研究分析了六种常见的头部姿势：右，左，上，下运动。基于小波尺度图的时频多分辨分析用于从信道状态信息值获得特征，沿着来自雷达信号的谱图用于头部运动检测。雷达和Wi-Fi信号的特征融合是通过最先进的DL模型进行的。VGG 16和InceptionV 3模型特征在雷达和Wi-Fi时频地图上分别训练，分别实现了83.33%和91.8%的高分类准确率。
The detection of head movement plays a crucial role in human–computer interaction systems. These systems depend on control signals to operate a range of assistive and augmented technologies, including wheelchairs for Quadriplegics, as well as virtual/augmented reality and assistive driving. Driver drowsiness detection and alert systems aided by head movement detection can prevent major accidents and save lives. Wearable devices, such as MagTrack consist of magnetic tags and magnetic eyeglasses clips and are intrusive. Vision-based systems suffer from ambient lighting, line of sight, and privacy issues. Contactless sensing has become an essential part of next-generation sensing and detection technologies. Wi-Fi and radar provide contactless sensing, however, in assistive driving they need to be inside enclosures or dashboards, which for all practical purposes in this article have been considered as through walls. In this study, we propose a contactless system to detect human head movement with and without walls. We used ultra-wideband (UWB) radar and Wi-Fi signals, leveraging machine and deep learning (DL) techniques. Our study analyzes the six common head gestures: right, left, up, and down movements. Time-frequency multiresolution analysis based on wavelet scalograms is used to obtain features from channel state information values, along with spectrograms from radar signals for head movement detection. Feature fusion of both radar and Wi-Fi signals is performed with state-of-the-art DL models. A high classification accuracy of 83.33% and 91.8% is achieved overall with the fusion of VGG16 and InceptionV3 model features trained on radar and Wi-Fi time–frequency maps with and without the walls, respectively.