Periodic Variance Maximization Using Generalized Eigenvalue Decomposition Applied to Remote Photoplethysmography Estimation

Richard Macwan, Serge Bobbia, Yannick Benezeth, Julien Dubois, Alamin Mansouri; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 1332-1340

A generic periodic variance maximization algorithm to extract periodic or quasi-periodic signals of unknown periods embedded into multi-channel temporal signal recordings is described in this paper. The algorithm combines the notion of maximizing a periodicity metric combined with the global optimization scheme to estimate the source periodic signal of an unknown period. The periodicity maximization is performed using Generalized Eigenvalue Decomposition (GEVD) and the global optimization is performed using tabu search. A case study of remote photoplethysmography signal estimation has been utilized to assess the performance of the method using videos from public databases UBFC-RPPG [??] and MMSE-HR [??]. The results confirm the improved performance over existing state of the art methods and the feasibility of the use of the method in a live scenario owing to its small execution time.