Scholarlink

Purpose <p>Remote photoplethysmography (rPPG) enables non-contact monitoring of vital signs using cameras but is highly sensitive to the selection of facial regions of interest (ROIs). Variations in signal quality across different facial areas and uneven light reflections can significantly degrade rPPG performance. To address this issue, this paper proposes an unsupervised facial ROI segmentation method that mitigates the impact of uneven illumination. By leveraging the spatial position and surface normal information of the three-dimensional (3D) facial contour, the proposed approach segments the facial ROI into multiple flat sub-ROIs, promoting uniform light reflection within each sub-region and improving rPPG signal quality.</p> Methods <p>First, a depth camera captures RGB-D facial video, from which a color point cloud of the face is generated. A super voxel clustering segmentation algorithm is then applied to divide the point cloud into multiple sub-ROIs, which are mapped to corresponding regions in the RGB image. Subsequently, an optical flow algorithm tracks each sub-ROI across video frames. Finally, rPPG signals are extracted from the sub-ROI sequences and fused to estimate heart rate (HR).</p> Results <p>The proposed method was evaluated on a self-collected RGB-D video dataset. It achieved a root mean square error (RMSE) of 6.32&#xa0;bpm under frontal illumination and 6.84&#xa0;bpm under 45-degree right-front illumination, both outperforming the comparison method.</p> Conclusion <p>The experimental results show that our ROI segmentation algorithm can effectively improve the accuracy of HR estimation and is robust to light unevenness.</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

Robust Heart Rate Measurement Against Uneven Illuminations Using an RGB-D Camera

  • 期刊论文

  • 发表日期：

    2025年6月7日

查看全文
  • Guang Yu,
  • Chenxi Yang,
  • Wenjin Wang,
  • Hongxiang Gao,
  • Zhijun Xiao,
  • Jianqing Li,
  • Chengyu Liu

摘要

Purpose

Remote photoplethysmography (rPPG) enables non-contact monitoring of vital signs using cameras but is highly sensitive to the selection of facial regions of interest (ROIs). Variations in signal quality across different facial areas and uneven light reflections can significantly degrade rPPG performance. To address this issue, this paper proposes an unsupervised facial ROI segmentation method that mitigates the impact of uneven illumination. By leveraging the spatial position and surface normal information of the three-dimensional (3D) facial contour, the proposed approach segments the facial ROI into multiple flat sub-ROIs, promoting uniform light reflection within each sub-region and improving rPPG signal quality.

Methods

First, a depth camera captures RGB-D facial video, from which a color point cloud of the face is generated. A super voxel clustering segmentation algorithm is then applied to divide the point cloud into multiple sub-ROIs, which are mapped to corresponding regions in the RGB image. Subsequently, an optical flow algorithm tracks each sub-ROI across video frames. Finally, rPPG signals are extracted from the sub-ROI sequences and fused to estimate heart rate (HR).

Results

The proposed method was evaluated on a self-collected RGB-D video dataset. It achieved a root mean square error (RMSE) of 6.32 bpm under frontal illumination and 6.84 bpm under 45-degree right-front illumination, both outperforming the comparison method.

Conclusion

The experimental results show that our ROI segmentation algorithm can effectively improve the accuracy of HR estimation and is robust to light unevenness.