Respiratory Rate Estimation by a Non-invasive Optical Method Based on Video Image Processing

Respiratory rate measurement and monitoring allow predicting adverse events on time. However, in clinical settings, it is performed with specialized devices that are always attached to the patient's body. The objective of this work was to estimate the instantaneous respiratory rate (IRR) and the respiratory movement (RM) using image processing techniques from a video. A smartphone recorded the videos, and the RM was estimated offline. An RM reference signal obtained in the same test with a thermal sensor around the nose and mouth was used to make comparisons with the computed signal. The remote optical estimation was evaluated on a pilot test with four volunteers in three different conditions of movement and breathing. We analyzed the correlation for the RM signals and the mean square error of the IRR between reference and estimated signals, obtaining a value of 0.737 and 0.091, respectively. These results are competitive with the ones founded in the bibliography. Moreover, the respiratory rate estimation has a precision of P = 0.98 and sensitivity S = 0.99, which makes our method superior. This outstanding performance is due to the proposed framework is robust in the presence of motion artifacts, even motion artifact that corrupts the reference signal from the thermal sensor.