Revisiting motion-based respiration measurement from videos

Area of Science:

• Biomedical Engineering
• Computer Vision
• Physiological Monitoring

Background:

• Contactless respiration rate monitoring utilizes video-based motion analysis to detect subtle chest or abdominal movements.
• Existing methods often rely on analyzing pixel intensity variations or movements within video frames.

Purpose of the Study:

• To evaluate video-based respiration monitoring techniques using a comprehensive benchmark with practical challenges.
• To compare the efficacy of pixel intensity variation versus pixel movement analysis for respiratory motion quantification.
• To assess the performance of conventional signal extraction methods against Convolutional Neural Network (CNN)-based approaches.

Main Methods:

• Development of a large video benchmark dataset encompassing six categories of practical challenges.
• Analysis of two key video properties: pixel intensity variation and pixel movement.
• Comparison of conventional signal extraction techniques (e.g., Zerophase Component Analysis) with CNN-based methods.

Main Results:

• Pixel movement analysis demonstrated superior quantification of respiratory motion compared to pixel intensity variation across various conditions.
• Conventional signal extraction methods, such as Zerophase Component Analysis, achieved higher performance than CNN-based methods.
• CNNs, despite data training, did not outperform simpler conventional approaches in extracting respiration signals.

Conclusions:

• Pixel movement is a more robust indicator for video-based respiration analysis than pixel intensity variation.
• Conventional signal processing techniques remain highly competitive and effective for contactless respiration monitoring.
• The study raises critical questions about the generalizability and necessity of CNNs for improving video-based physiological signal measurements.