Reduction of Motion Artifacts From Remote Photoplethysmography Using Adaptive Noise Cancellation and Modified HSI Model

Abstract

Remote photoplethysmography (rPPG) is a method to measure cardiac activities without any contact sensors. Non-contact sensors include radar, laser, and digital cameras, and there have been wide developments regarding the measurement of rPPG signals using continuous face frames. However, non-contact sensors are quite sensitive to the subject’s motion, which causes motion artifacts. In this paper, two hypotheses are proposed: a) the motion artifacts are caused by unevenly reflected light due to the curvature of the subject’s face; and b) melanin and residuals in the continuous face frames are time-varying values whenever the subject’s movement is triggered. Adaptive noise cancellation based on recursive least square (ANS based on RLS) using the Lambert-Beer law and the hue–saturation–intensity (HSI) model were applied. The former is used for skin modeling, and the latter is used to reduce noises derived by the curvature of the face. Furthermore, the proposed algorithm is directly applied to two-dimensional continuous face frames and results in the rPPG signal and rPPG image, respectively. To evaluate proposed algorithm, two different experiments (e.g., static and dynamic situation) were conducted. Furthermore, in a study with 15 participants, the performances of heart rate estimation and heart rate variability (HRV) were evaluated by comparing the proposed method with previously developed methods. The results showed that a) the artifacts derived by head movement are efficiently removed, compared to previous methods; and b) rPPG images describing the spread of facial blood flow are acquired in real-time.

Keywords

• Adaptive noise cancellation (ANC)
• heart rate variability (HRV)
• HSI color model
• photoplethysmography (PPG)
• recursive least square (RLS)
• remote sensing