Scholarlink

This paper introduces a comprehensive framework for detecting facial regions of interest (ROIs) from video data and predicting photoplethysmogram (PPG) signals through a deep learning approach. The proposed system leverages a YOLO v8 model for efficient, real-time face detection, combined with advanced image processing techniques to isolate key facial regions such as the forehead, cheeks, and nose. The extracted ROIs undergo preprocessing steps, including normalization and resizing for uniformity. Temporal features are derived using a sliding window approach, which calculates statistical measures to capture the time-dependent characteristics of the PPG signal. These temporal features are then processed by a neural network model, which predicts PPG outputs in a three-channel configuration. The model is trained using mean squared error loss and evaluated with mean absolute error, showcasing the pipeline’s effectiveness in estimating PPG signals non-invasively. This approach highlights the potential of integrated computer vision and deep learning techniques for non-contact physiological monitoring.

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

A Deep Learning Framework for Predicting Temporal PPG Signals Through Facial Region Extraction with YOLO

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ayesha Sadiqqa,
  • Shi-Jinn Horng,
  • Chandramohan Reddy Poreddy

摘要

This paper introduces a comprehensive framework for detecting facial regions of interest (ROIs) from video data and predicting photoplethysmogram (PPG) signals through a deep learning approach. The proposed system leverages a YOLO v8 model for efficient, real-time face detection, combined with advanced image processing techniques to isolate key facial regions such as the forehead, cheeks, and nose. The extracted ROIs undergo preprocessing steps, including normalization and resizing for uniformity. Temporal features are derived using a sliding window approach, which calculates statistical measures to capture the time-dependent characteristics of the PPG signal. These temporal features are then processed by a neural network model, which predicts PPG outputs in a three-channel configuration. The model is trained using mean squared error loss and evaluated with mean absolute error, showcasing the pipeline’s effectiveness in estimating PPG signals non-invasively. This approach highlights the potential of integrated computer vision and deep learning techniques for non-contact physiological monitoring.