OPOIRES: A robust non-contact respiratory rate extraction based on optimal points-of-interest selection from an RGB camera

• Published in: Biomedical signal processing and control Vol. 86; p. 105193
• Main Authors: Deo Mehta, Arya, Sharma, Hemant
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2023
• Subjects: Camera; Covid-19; HAI; iPPG; Respiratory rate; Vital signs; HAI; Respiratory rate; Camera; Vital signs; Covid-19; iPPG
• DOI: 10.1016/j.bspc.2023.105193
• ISSN: 1746-8094

•Explored phase and intensity coherency in respiratory-induced motion (RIM) across frontal region using RGB camera.•Conceptualized and tracked unique points of interest (POIs) in the frontal region for RIM monitoring.•Developed RWD mechanism to select informative POIs for accurate respiratory rate (RR) estimation.•Demonstrated high accuracy and robustness of the proposed framework across diverse conditions through experiments on large datasets. The conventional approaches to monitoring respiratory rate (RR) require wearable sensors, which are cumbersome and cause discomfort. Advancements in RGB cameras and computer vision algorithms have led to a possible technological solution providing a comfortable, feasible, and cost-effective method to monitor RR. Lately, researchers have made significant attempts to estimate RR through contactless means using an RGB camera. In light of this, we propose a new but robust algorithm using Optimal Points-Of-Interest based RESpiratory rate estimation (OPOIRES). The algorithm is developed after understanding the nature of the influence of respiratory-induced motion (RIM) in the frontal region of a human subject, such as the torso and shoulders. In the process, it is observed that the phase and amplitude of RIM vary with the region of observation. To track these changes, we conceptualized a local points-of-interest (POIs)-based approach instead of a region of interest and used a two-stage process to select a few POIs possessing the best RIM characteristics, which are eventually used to estimate RR. The performance of the algorithm is analyzed over two large publicly available datasets, namely DEAP and COHFACE, for multiple window lengths, including 20 s, 30 s, and 60 s. In experimental analysis, the technique is observed to be achieving high accuracy in RR estimation and possessing tolerance to illumination variation to a great extent. In comparison, the proposed technique outperformed the existing methods in all the data classes. The results highlight the potential of the OPOIRES framework for contactless RR monitoring through consumer-grade cameras such as webcams and smartphones.