Motion Robust Imaging Ballistocardiography Study With Canonical Correlation Analysis

Imaging ballistocardiography(iBCG)is a video-based non-contact heart rate measurement technology.It extracts the HR from the subject’s head in the video by processing the weak mechanical movement caused by the beating of the heart.This technology has attracted much attention from researchers due to its simple operation and low cost.However,the iBCG measurements can be easily distorted due to the rigid motions caused by voluntary movements or the nonrigid motions resulted from facial expressions.In this thesis,we propose a novel iBCG method,called robust iBCG(RiBCG),to suppress motion artifacts of iBCG measurements with a two-step canonical correlation analysis(CCA): First,considering the correlation between the horizontal and vertical traces in each ROI,the first CCA is taken to separately remove the rigid motion artifacts of raw iBCG measurements with respect to the horizontal traces in each ROI.The principal component analysis(PCA)is then used to further reduce the dimensionality of rigid-motion-free iBCG signals.Then,according to the spatial correlation of the HR in the two ROIs,the CCA is applied again to extract the common components of the two sets of principal components obtained in the previous step,which is to suppress non-rigid motion artifacts with low spatial correlation.Finally,the pulse signal is selected to calculate the heart rate.Besides,an improved version of RiBCG,termed as RiBCG-C,is also proposed to reduce the HR outliers considering the continuity of HR variations.The test results on the public databases UBFC-RPPG and COHFACE show that the proposed RiBCG-C method,compared with several typical video-based heart rate measurement methods,achieves overall the best performance,and the Pearson correlation coefficient on UBFC-RPPG and COHFACE databases reached 0.96 and 0.85,respectively.The study provides an effective scheme for iBCG measurements under realistic application scenarios.