Parameter-free adaptive step-size multiobjective optimization applied to remote photoplethysmography

Richard Macwan (1) , Yannick Benezeth (1) , Keisuke Nakamura (2) , Randy Gomez (2) , Yadong Wu (3) , Alamin Mansouri (4)

1 Le2i - Laboratoire d'Electronique, d'Informatique et d'Image [EA 7508]
2 Honda Research Institute
3 SWUST - Southwest University of Science and Technology [Mianyang]
4 Le2i - Laboratoire Electronique, Informatique et Image [UMR6306]

In this work, we propose to reformulate the objective function of Independent Component Analysis (ICA) to make it a better posed problem in the context of Remote photoplethysmography (rPPG). In recent previous works, linear combination coefficients of RGB channels are estimated maximizing the non-Gaussianity of ICA output components. However, in the context of rPPG a priori knowledge of the pulse signal can be incorporated into the component extraction algorithm. To this end, the contrast function of regular ICA is extended with a measure of periodicity formulated using autocorrelation. This novel semi-blind source extraction method for measuring rPPG has the interesting property of being free from manual parameter adjustment. The tedious selection of the step-size parameter in the gradient-ascent algorithm has been advantageously replaced by an adaptive step size. Our method has been validated against our large in-house video database UBFC-RPPG.

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]; Traitement du signal et de l'image [eess.SP]

Richard Macwan, Yannick Benezeth, Keisuke Nakamura, Randy Gomez, Yadong Wu, et al.. Parameter-free adaptive step-size multiobjective optimization applied to remote photoplethysmography. IEEE International Conference on Biomedical and Health Informatics, 2018, Las Vegas, United States. ⟨hal-01678241⟩