Non-contact heart rate detection based on pixel time series classification

Non-contact heart rate (HR) detection is a new technique developed in recent years, which uses face videos captured by ordinary cameras for heart rate signal estimation, and the implementation of this technique can be divided into two major parts: signal extraction and signal analysis. This paper classifies each pixel timing by using a simplified nearest neighbor classifier and optimizes the signal extraction results by morphological filtering. In the signal analysis stage, this paper uses principal component analysis (PCA) to pre-process the extracted signal set to select the three principal elements that are most likely to contain the heart rate signal, and then uses independent component analysis (ICA) to separate the three principal elements blindly, and then determines the final heart rate signal by fast Fourier transform (FFT). In this paper, the heart rate signal obtained from the above method using the laptop's own camera is compared with that of a finger clip pulse oximeter, and it is found that the algorithm used in this paper is used for non-contact heart rate detection in the presence of ambient noise and slight motion effects.