Contactless Heart Rate Estimation From Facial Video Using Skin Detection and Multi-resolution Analysis

This paper introduces a remote Photoplethysmography (rPPG), which is used to estimate human heart rate withoutany physical contact, has been extensively applied in multiple fields like medical diagnosis, analysis of humanemotions, rehabilitation training programs, biometric, and fitness assessments. The rPPG signals are usually ex-tracted from facial videos. However, it is still a challenging task due to several contributing factors, e.g., variationin skin tone, lighting condition, and subject’s motion. Accordingly, in this work, a novel approach based on deeplearning skin detection method and the discrete wavelet transform (DWT) is employed to precisely estimate heartrate from facial videos. In the proposed method, by implementing the DWT, the signal is decomposed into approx-imations and details parts thereby it helps in analyzing it at different frequency bands with different resolutions.The results derived from the experiments show that our proposed method outperforms the state-of-the-art methodson the UBFC-RPPG database.