Video-based non-contact blood oxygen detection method

Abstract

The invention provides a video-based non-contact blood oxygen detection method, and aims to accurately predict blood oxygen saturation (SpO2) by shooting a hand video and combining an image processing technology and a machine learning algorithm. The method comprises the following steps of: firstly, segmenting a hand region by using a maximum between-class variance method, and optimizing a region of interest (ROI) through corrosion and expansion operations; then, RGB signals in the video frames are extracted through a spatial averaging method, high-frequency noise is removed through an eight-order Butterworth low-pass filter, and direct-current components are extracted; then, principal component analysis (PCA) is used to extract a first principal component from red-green, red-blue, and green-blue channel combinations to obtain an AC signal and calculate a standard deviation thereof as an AC feature. And finally, training and predicting the oxyhemoglobin saturation by using a support vector regression (SVR) model and combining the direct-current features and the alternating-current features. According to the method, the generalization ability of the model is improved through five-fold cross validation, and finally high-precision blood oxygen detection is achieved.

Images