A Novel Algorithm for Remote Photoplethysmography: Spatial Subspace Rotation

Remote photoplethysmography (rPPG) enables contactless monitoring of the blood volume pulse using a regular camera. Recent research focused on improved motion robustness, but the proposed blind source separation techniques (BSS) in RGB color space show limited success. We present an analysis of the motion problem, from which far superior chrominance-based methods emerge. For a population of 117 stationary subjects, we show our methods to perform in 92% good agreement ( ±1.96σ) with contact PPG, with RMSE and standard deviation both a factor of 2 better than BSS-based methods. In a fitness setting using a simple spectral peak detector, the obtained pulse-rate for modest motion (bike) improves from 79% to 98% correct, and for vigorous motion (stepping) from less than 11% to more than 48% correct. We expect the greatly improved robustness to considerably widen the application scope of the technology.

Remote sensing of the reflectance photoplethysmogram using a video camera typically positioned 1 m away from the patient's face is a promising method for monitoring the vital signs of patients without attaching any electrodes or sensors to them. Most of the papers in the literature on non-contact vital sign monitoring report results on human volunteers in controlled environments. We have been able to obtain estimates of heart rate and respiratory rate and preliminary results on changes in oxygen saturation from double-monitored patients undergoing haemodialysis in the Oxford Kidney Unit. To achieve this, we have devised a novel method of cancelling out aliased frequency components caused by artificial light flicker, using auto-regressive (AR) modelling and pole cancellation. Secondly, we have been able to construct accurate maps of the spatial distribution of heart rate and respiratory rate information from the coefficients of the AR model. In stable sections with minimal patient motion, the mean absolute error between the camera-derived estimate of heart rate and the reference value from a pulse oximeter is similar to the mean absolute error between two pulse oximeter measurements at different sites (finger and earlobe). The activities of daily living affect the respiratory rate, but the camera-derived estimates of this parameter are at least as accurate as those derived from a thoracic expansion sensor (chest belt). During a period of obstructive sleep apnoea, we tracked changes in oxygen saturation using the ratio of normalized reflectance changes in two colour channels (red and blue), but this required calibration against the reference data from a pulse oximeter.

Using a time-lapse image acquired from a CCD camera, we developed a non-contact and non-invasive device, which could measure both the respiratory and pulse rate simultaneously. The time-lapse image of a part of the subject's skin was consecutively captured, and the changes in the average image brightness of the region of interest (ROI) were measured for 30s. The brightness data were processed by a series of operations of interpolation as follows a first-order derivative, a low pass filter of 2 Hz, and a sixth-order auto-regressive (AR) spectral analysis. Fourteen sound and healthy female subjects (22-27 years of age) participated in the experiments. Each subject was told to keep a relaxed seating posture with no physical restriction. At the same time, heart rate was measured by a pulse oximeter and respiratory rate was measured by a thermistor placed at the external naris. Using AR spectral analysis, two clear peaks could be detected at approximately 0.3 and 1.2 Hz. The peaks were thought to correspond to the respiratory rate and the heart rate. Correlation coefficients of 0.90 and 0.93 were obtained for the measurement of heart rate and respiratory rate, respectively.