Improved Heart Rate Estimation From Facial Videos Using Hair Detection and Majority Vote in Subintervals

• Main Authors: Sunkom, Panupong, Worasawate, Denchai, Srisurangkul, Chadchai, Nakayama, Minoru
• Format: Conference Proceeding
• Language: English
• Subjects: Computer science; Estimation; Forehead; Fourier transforms; Hair; Hair detection; Heart rate; Heart rate estimation; Photoplethysmography; Remote photoplethysmography
• ISSN: 2642-6579
• DOI: 10.1109/JCSSE58229.2023.10201996

Remote photoplethysmography (rPPG) is a non-contact method that can be used to estimate heart rate (HR) from facial video. Many regions of interest (ROI) on the face were suggested for the rPPG signal extraction, such as the forehead and cheek. However, for the forehead ROI, some parts of the skin area may be covered by hair. This could lead to incorrect rPPG signals and HR. This paper proposes a method to improve the quality of forehead ROI by detecting and removing parts covered by hair based on the green color signal extracted from the forehead area. Any change in ambient light could introduce spikes of spurious frequencies in the interested interval. These spurious frequencies might be stronger than extracted rPPG signals. To overcome these spikes, subintervals are considered. The Short Time Fourier Transform (STFT) is applied to the rPPG signal of the interested interval to obtain HR for each subinterval. The representative HR for the interested interval is selected by majority vote. The estimated HR on the interested interval is then computed based on the representative HR. These experiments were performed on a public dataset, UBFC-RPPG, and the results show that the mean absolute error (MAE) of HR is improved by the proposed method.