BodySensor

Real-time tracking of electricians in distribution rooms is essential for ensuring operational safety. Traditional GPS-based methods, however, are ineffective in such environments due to complex non-line-of-sight (NLOS) conditions caused by dense cabinets and thick walls that obstruct satellite signals. Existing solutions, such as video-based systems, are prone to inaccuracies due to NLOS effects, while wearable devices often prove inconvenient for workers. To address these challenges, we propose BodySensor, a contactless system that utilizes a single commercial off-the-shelf (COTS) ultra-wideband (UWB) radar for real-time, fine-grained indoor tracking in distribution rooms. Our approach incorporates the Native Resolution Vision Transformer (NaViT), a Vision Transformer variant, to identify human positions by analyzing nonlinear multipath signals. Furthermore, we introduce an RF-to-real position mapping to enhance positioning accuracy within detected intervals. Extensive experiments conducted in three distinct environments, encompassing over 30 hours of data, demonstrate that BodySensor achieves a coarse localization accuracy of up to $99.8\%$, with a mean tracking error of approximately 6.2 cm under line-of-sight (LOS) conditions and 10.9 cm under full NLOS conditions. The system maintains consistent accuracy regardless of whether the subject is stationary or in motion, underscoring its robustness and practicality. BodySensor exemplifies the potential of radar-based solutions for smart power systems, providing a reliable and non-intrusive alternative for enhancing worker safety and operational efficiency.

配电室内电工的实时追踪对于保障作业安全至关重要。然而，传统的基于全球定位系统（Global Positioning System, GPS）的方法在这类环境中效果欠佳，原因在于密集机柜与厚重墙体所营造的复杂非视距（non-line-of-sight, NLOS）环境会遮挡卫星信号。现有解决方案（如基于视频的系统）易受NLOS效应影响而产生定位误差，而可穿戴设备往往会给作业人员带来使用不便。为应对上述挑战，本文提出BodySensor系统：这是一种非接触式系统，借助单台商用现货（commercial off-the-shelf, COTS）超宽带（ultra-wideband, UWB）雷达，实现配电室内的实时、细粒度室内追踪。我们的方案采用了原生分辨率视觉Transformer（Native Resolution Vision Transformer, NaViT）——一种视觉Transformer变体——通过分析非线性多径信号来识别人体位置。此外，我们提出了射频-真实位置映射方案，以提升检测区间内的定位精度。研究团队在三种不同环境中开展了大规模实验，总计涵盖超过30小时的实验数据，结果表明BodySensor的粗定位准确率最高可达99.8%：在视距（line-of-sight, LOS）环境下平均追踪误差约为6.2厘米，在全非视距（NLOS）环境下则为10.9厘米。无论被测人员处于静止还是运动状态，该系统均能保持稳定的定位精度，凸显了其鲁棒性与实用性。BodySensor验证了雷达方案在智能电力系统中的应用潜力，可为提升作业人员安全与运营效率提供一种可靠且无接触的备选方案。

搜集汇总

背景与挑战

背景概述

BodySensor数据集专注于利用单台商用超宽带雷达实现配电室电工的实时非接触式室内跟踪，以解决传统GPS和视频方法在非视距环境下的不足。系统通过Native Resolution Vision Transformer分析多径信号，并结合RF到真实位置映射，在超过30小时的实验中，在视距条件下平均跟踪误差约6.2厘米，非视距条件下约10.9厘米，准确率高达99.8%，体现了其高精度和鲁棒性，适用于智能电力系统以提升工人安全和操作效率。