Video Based Noncontact Heart Rate Estimation with Scale Invariant Feature Transform and Nonlinear Mode Decomposition

The average heart rate (HR) can be determined from facial videos using non-contact remote photoplethysmog-raphy (rPPG). However, practical applications continue to be constrained by issues such as illumination variations, non-rigid head movement and face motion. This paper describes a complete system structure for remotely predicting heart rate from face video under a variety of apparent motion situations. The proposed method is based on automated face tracking and nonlinear mode decomposition (NMD) along with scale invariant feature transform (SIFT) features from the current and past frames to adjust the face of the current frame to the location of the face in the prior frame using the affine transform (AT). The experimentation is carried out on COHFACE and DROZY datasets. Heart rate estimated with proposed method is compared with the heart rate estimated using finger blood volume pulse sensor. This comparision is carried out with Bland-Altman and correlation analysis. The proposed method when compared to other methods utilizing comparable datasets outperforms in terms of mean absolute error (MAE), root mean square error (RMSE), standard deviation (SD), and correlation coefficient (CC).