Video-Based Breathing Rate Monitoring in Sleeping Subjects

This paper addresses the challenge of detecting the breathing cessation in sleeping subjects, via breathing pattern monitoring at a distance and under "night-light" conditions. We investigate a near-infrared video-based approach to estimate the breathing rate, based on chest or back movements. A body pose estimation algorithm and the Lucas-Kanade optical flow method are combined to automatically detect the Region of Interest (ROI) represented by a grid of points. The movement of the ROI is then translated into the frequency of respiratory events. We used a dataset with 28 near-infrared videos, as well as 11 videos of subject uncovered and partially covered by blankets. We compared the breathing rate measurements provided by a wearable device with the ones estimated by the video-based approach. A linear correlation analysis of both measurements resulted in a coefficient of determination of 0.925, and accuracy of 99.70% for the first dataset, and 0.873 and 88.95% for the second dataset, respectively. The ultimate application is to detect abnormalities in breathing and health emergencies in environments such as homeless shelters.