Related Experiment Video
Updated: Jun 19, 2025
12:39
A Methodology for Capturing Joint Visual Attention Using Mobile Eye-Trackers
Published on: January 18, 2020
7.6K
SaccpaNet: A Separable Atrous Convolution- Based Cascade Pyramid Attention Network to Estimate Body Landmarks Using
IEEE Journal of Biomedical and Health Informatics
|July 23, 2024
Summary
This study introduces SaccpaNet, a deep learning system using depth cameras for accurate sleep posture classification at home. The model effectively handles blanket interference, achieving high accuracy for reliable sleep monitoring.
Area of Science:
- Computer Vision and Machine Learning
- Biomedical Engineering
- Sleep Science
Background:
- Standard polysomnography sleep posture assessment accuracy can be affected by unfamiliar laboratory environments.
- There is a need for reliable sleep posture monitoring systems suitable for home or community use.
Purpose of the Study:
- To develop a depth camera-based system for sleep posture monitoring and classification outside of clinical settings.
- To create a deep learning model, SaccpaNet, capable of accurately classifying sleep postures while accounting for blanket interference.
Main Methods:
- Developed SaccpaNet, incorporating a joint coordinate estimation network (JCE) and sleep posture classification network (SPC) with a pyramidal attention structure.
- Utilized cross-modal pretraining with RGB images (COCO whole body dataset) and trained/tested on depth images from 150 participants across seven sleep postures and four blanket conditions.
- Implemented data augmentation techniques, including intra-class mix-up and overlaid flip-cut, for enhanced robustness against blanket interference (PhD-ART).
Main Results:
- Achieved an average precision of 0.652 for estimated joint coordinates (PCK@0.1), demonstrating robust pose estimation.
- Attained high sleep posture classification accuracy: F1-scores of 0.885 for 7-class and 0.940 for 6-class classification.
- The system showed significant resistance to blanket interference, with a spread difference of only 2.5%.
Conclusions:
- The developed SaccpaNet system provides accurate and robust sleep posture monitoring and classification using depth cameras in non-clinical environments.
- The system's ability to handle blanket interference makes it a promising tool for unobtrusive, long-term sleep studies.
- This technology can facilitate more naturalistic sleep assessments, potentially improving the understanding of sleep-related behaviors and conditions.
