Related Experiment Video
Updated: Jun 19, 2025
11:34
High-resolution, High-speed, Three-dimensional Video Imaging with Digital Fringe Projection Techniques
Published on: December 3, 2013
15.6K
3D DenseNet with temporal transition layer for heart rate estimation from real-life RGB videos
Summary
This study introduces a 3D DenseNet model for accurate heart rate estimation from videos, outperforming traditional methods in real-world scenarios. The deep learning approach achieves reliable results even without predefined regions of interest.
Area of Science:
- Biomedical Engineering
- Computer Vision
- Machine Learning
Background:
- Deep learning excels at heart rate estimation in controlled settings.
- Performance varies in uncontrolled scenarios due to datasets and model architectures.
Purpose of the Study:
- Develop a deep learning model for accurate heart rate estimation from RGB videos.
- Utilize 3D convolutional neural networks (3DCNN) to capture temporal and spatial features.
- Achieve accurate heart rate estimation without a predefined region of interest (ROI).
Main Methods:
- Proposed a 3D DenseNet model with a 3D temporal transition layer.
- Trained and tested the model on a large-scale dataset of real-life, hospital-like videos.
- Leveraged 3DCNN for feature extraction from video data.
Main Results:
- Achieved a root mean square error (RMSE) of 8.68 BPM and a mean absolute error (MAE) of 3.34 BPM on the custom dataset.
- Demonstrated superior performance compared to state-of-the-art models on the VIPL-HR public dataset.
Conclusions:
- The developed 3D DenseNet model provides accurate heart rate estimation in less controlled video environments.
- The model's ability to extract spatio-temporal features enhances its robustness and generalizability.
- This approach offers a promising non-invasive method for continuous heart rate monitoring.
