Motion-robust anterior–posterior imaging ballistocardiography for non-contact heart rate measurements

Abstract

Imaging ballistocardiography (iBCG) is a novel technique that utilizes video-based technology to measure heart rate (HR). This innovative method is based on the detection of subtle mechanical head movements that are caused by heartbeats and does not require direct physical contact with the body. However, the iBCG signals can be easily contaminated by rigid and nonrigid motion artifacts. In this paper, we propose a novel method for iBCG motion artifact removal (MAR). First, anterior–posterior (Z-axis) signals are reconstructed as the raw iBCG signals from two facial regions of interest (ROIs) according to the law of perspective. Next, the adaptive filtering technique is utilized to remove rigid motion artifacts from raw iBCG signals with respect to the motion reference of the X-axis traces. The rigid-motion-free iBCG signals in each ROI are further dimensionally compressed via principal component analysis (PCA). Then, the principle components (PCs) from the two regions of interest (ROIs) are subjected to canonical correlation analysis (CCA) to remove residual non-rigid motion artifacts that have low spatial correlations. Finally, the target HR value is identified by selecting the canonical variates (CVs) with the highest power spectrum amplitude. Experimental results on two publicly available databases and a in-house real-world database demonstrate that the proposed method achieves the best performance compared to existing iBCG methods during most cases, especially in the presence of motion artifacts. The study provides a feasible scheme for practical applications of iBCG HR measurements.