Research On Emotion Recognition Method Based On MmWave Non-contact Heart Rate Variability Monitoring

With the accelerated pace of life and the social background of repeated epidemics,people are facing multiple pressures such as life and work,and more and more people are experiencing negative emotional disorders.Emotion plays a key role in interpersonal emotional communication.Currently,research on emotion recognition is mainly divided into two categories: based on non physiological signals and based on physiological signals.However,the former is extremely susceptible to subjective concealment,while the latter is not subject to subjective influence,but most of the information is collected through wearable devices,which not only causes adverse experiences for users,but also affects their emotional state.In contrast,the non-contact acquisition(physiological signal)method shows its advantages and prospects.In recent years,experts have found that heart rate variability analysis of ECG signals has great value in emotional recognition.This article focuses on the method of non-contact acquisition of ECG signals using millimeter wave radar signals to achieve the purpose of using heart rate variability features to identify emotions.The main research work can be summarized as follows:Firstly,a respiration and heartbeat detection model based on FMCW millimeter wave radar is proposed.Starting from the characteristics of ECG signals,the relationship between chest vibration caused by respiration and heartbeat and the change amount of radar IF signals is established.A moving target detection algorithm is used to determine the position of the target,extract temporal phase information through phase unwrapping,phase subtraction,and normalization.A bandpass filter is used to separate the respiration and heartbeat signals,and the heartbeat signals are extracted using variational mode decomposition for subsequent heart rate variability analysis.Secondly,a heartbeat HRV feature extraction method and an emotion recognition algorithm are proposed.The peak detection algorithm is used to extract the heart rate R wave sequence(IBI sequence),and then statistical methods are used to calculate the heart rate variability characteristics in the time domain,frequency domain,and nonlinear domain,including 13 dimensional heart rate variability characteristics such as HR,SDNN,RMSSD,SDSD,LF,and HF.The above characteristics are composed of emotional feature data and input into the SVM classification algorithm for classification and recognition.Finally,a non-contact emotion recognition system mm HRV based on support vector machines is proposed.Using the video editing method to induce four types of emotions,a self collected emotional dataset of user independent models was established,with more than 50 samples of each emotion.The performance of the mm HRV system was evaluated using the self collected emotional dataset from different classifiers,different distances,and different angles.The results show that the self collected data sets of eight users have achieved relatively ideal recognition results,and the average recognition accuracy of the mm HRV system in the experimental scene reaches 95.60%.