Scholarlink

<p>Ensuring the safety of personnel operating within distribution rooms is critical for maintaining electrical system reliability. Accurate identification and monitoring of human activities are foundational to achieving this goal. However, conventional detection methods in power environments face challenges such as equipment complexity, electromagnetic interference, and limited data granularity, restricting their practical utility. This paper introduces WiPowerSys, a novel framework for skeleton-based human pose estimation in device-free distribution rooms using channel state information (CSI). We develop a custom CSI sensor based on the ESP32 platform to capture activity-induced signal variations, complemented by a Kinect v2 camera to annotate skeletal joint data. The framework employs a hybrid deep learning model integrating residual networks and bidirectional gated recurrent units to uncover intricate correlations between power system operations and CSI dynamics. A convolutional block attention module further enhances the model’s focus on critical local features. Pose estimation accuracy is evaluated by computing the Euclidean distance between ground-truth joint coordinates from the Kinect v2 and model predictions. Experiment results demonstrate that WiPowerSys achieves precise skeletal tracking in electromagnetically noisy power scenarios, with 88.71% of predictions exhibiting an Euclidean distance below 20&#xa0;mm. This approach not only captures subtle activity distinctions but also advances the efficacy of non-intrusive safety monitoring in distribution rooms.</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

WiPowerSys: A Framework Using CSI Signal for Skeleton-Based Human Pose Estimation in Distribution Room

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • First Jing Chen,
  • Second Xiangxi Li,
  • Third Yiyuan Liang,
  • Fourth Cunyi Yin,
  • Fifth Hao Jiang,
  • Sixth Deying Chen

摘要

Ensuring the safety of personnel operating within distribution rooms is critical for maintaining electrical system reliability. Accurate identification and monitoring of human activities are foundational to achieving this goal. However, conventional detection methods in power environments face challenges such as equipment complexity, electromagnetic interference, and limited data granularity, restricting their practical utility. This paper introduces WiPowerSys, a novel framework for skeleton-based human pose estimation in device-free distribution rooms using channel state information (CSI). We develop a custom CSI sensor based on the ESP32 platform to capture activity-induced signal variations, complemented by a Kinect v2 camera to annotate skeletal joint data. The framework employs a hybrid deep learning model integrating residual networks and bidirectional gated recurrent units to uncover intricate correlations between power system operations and CSI dynamics. A convolutional block attention module further enhances the model’s focus on critical local features. Pose estimation accuracy is evaluated by computing the Euclidean distance between ground-truth joint coordinates from the Kinect v2 and model predictions. Experiment results demonstrate that WiPowerSys achieves precise skeletal tracking in electromagnetically noisy power scenarios, with 88.71% of predictions exhibiting an Euclidean distance below 20 mm. This approach not only captures subtle activity distinctions but also advances the efficacy of non-intrusive safety monitoring in distribution rooms.