| Psychological stress is an innate natural reaction of human beings,and when it is severe,it will have a negative impact on physical and mental health.With the promotion of mental health knowledge,more and more people begin to pay attention to their own psychological stress.Current psychological stress detection methods include electroencephalogram detection,hormone level measurement,psychological scale test and other methods.These methods have high accuracy but have disadvantages such as complicated operation,need for professional and equipment support,and inability to perform home detection.Remote Photoplethysmography(r PPG)is a non-contact physiological index detection method that uses a color camera as a sensor to measure Blood Volume Pulse(BVP)signals,which can be processed and analyzed to obtain heart rate and heart rate variability(HRV)and other physiological indicators,suitable for remote detection of psychological stress and estimation of physiological indicators.Therefore,this thesis designs a remote detection method of psychological stress based on r PPG,which is run on a computer equipped with a camera to detect whether the user has a psychological stress problem.The experimental results on the UBFC-Phys dataset show that the method has a good classification effect on psychological stress.The main work of this thesis is as follows:(1)A BVP signal extraction and enhancement method is designed.First,the distribution of BVP signal intensity in the face area is studied and the region of interest is selected to ensure that the signal has a higher signal-to-noise ratio.Secondly,the face detection model based on Res Net-18 and the kernel correlation function KCF tracking algorithm are used for face positioning,and the skin segmentation is performed based on the Gaussian skin color model to reduce the introduction of noise.Finally,the BVP signal is extracted based on the chromaticity model,and a signal enhancement model based on gradient boosting regression is proposed to enhance the BVP signal,which provides support for the subsequent remote detection of psychological stress.(2)A heart rate remote estimation model is constructed.First,in response to the lack of an open source head-fixed heart rate dataset,this paper constructed a UJN-r PPG dataset containing 45,000 frames of images and heart rate information in a laboratory environment.Second,a lightweight heart rate remote estimation method based on spatio-temporal feature maps is proposed.Spatio-temporal feature maps are generated from the BVP-enhanced signal as input to a convolutional neural network to obtain accurate heart rate estimates as input features for subsequent models.(3)A remote detection model of psychological stress is constructed.A psychological stress classification model based on Ada Boost-SVM was proposed,and the NN interval of BVP enhanced signal was calculated and analyzed to obtain HRV features.After feature selection,it was fused with heart rate features as the input of the model.The experimental results on the remote psychological stress detection data set UBFC-Phys show that the psychological pressure classification accuracy of the constructed model reaches 77.29%,showing good application value.(4)Design and implement a mental health detection system,which is based on the MVC architecture and has the functions of BVP signal import,heart rate estimation,psychological stress detection and report export,etc.,and realizes the efficient management and application of BVP signal,heart rate information and psychological stress information. |
