Non-contact heart rate detection by combining empirical mode decomposition and permutation entropy under non-cooperative face shake

Abstract

Video-based non-contact heart rate detection can be easily affected by factors such as face shake and shooting environment; thus, effectively extracting the blood volume pulse signal is difficult. Therefore, a video-based face-shake-resistant heart rate detection method was proposed in this paper to mediate this problem. First, the face region that was selected through the multi-task convolution neural networks was used to correct the tilt angle and obtain the face image sequence. The face image sequence possessed approximately the same skin color information. Afterward, empirical mode decomposition and permutation entropy were combined, and the initial position of the signal was determined according to the randomness of the intrinsic mode function component to denoise and reconstruct the blood volume pulse signal. Finally, spectral analysis was implemented for the reconstructed signal to compute the heart rate value. The experimental results showed that the proposed method was highly consistent with the measurement result of the pulse oximeter; moreover, the proposed method showed good stability and accuracy for human heart rate detection.