Scholarlink

<p>Surveillance systems are pivotal in public safety, yet they often fail under adverse conditions such as darkness, occlusion, or deliberate tampering. Millimeter-wave (mmWave) radar, operating independently of light, can capture human motion even when cameras are obstructed. This paper introduces R2P2V, a two-stage framework that reconstructs surveillance videos from mmWave radar signals when cameras fail. The first stage estimates pose heatmaps from raw mmWave radar data, while the second stage employs a lightweight pose-guided generator to synthesize RGB frames using deep radar features and pre-acquired visual references. Experiments on the HuPR dataset demonstrate that R2P2V improves image quality and human-shape consistency over baseline methods, particularly in low-light and occluded scenarios. The code of this paper has been released at: <a href="https://github.com/cocein/R2P2V.">https://github.com/cocein/R2P2V.</a></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

Millimeter-wave radar-assisted skeleton-guided video reconstruction for surveillance systems

  • 期刊论文

  • 发表日期：

    2026年4月21日

查看全文
  • Wenjie Leng,
  • Bing Luo,
  • Mao Dai,
  • Li Xu

摘要

Surveillance systems are pivotal in public safety, yet they often fail under adverse conditions such as darkness, occlusion, or deliberate tampering. Millimeter-wave (mmWave) radar, operating independently of light, can capture human motion even when cameras are obstructed. This paper introduces R2P2V, a two-stage framework that reconstructs surveillance videos from mmWave radar signals when cameras fail. The first stage estimates pose heatmaps from raw mmWave radar data, while the second stage employs a lightweight pose-guided generator to synthesize RGB frames using deep radar features and pre-acquired visual references. Experiments on the HuPR dataset demonstrate that R2P2V improves image quality and human-shape consistency over baseline methods, particularly in low-light and occluded scenarios. The code of this paper has been released at: https://github.com/cocein/R2P2V.