| At present,the researches on emotion recognition based on heart rate variability(HRV)are more and more extensive and gradually maturing,but its HRV parameters are obtained by contact method.However,in practical applications,such as in the cabin,square,etc.contact-type physiological signal collection can cause inconvenience to passengers if we want to monitor the subjects' emotion using HRV parameters without disturbing their normal activities.To this end,the camera is used in this paper to capture facial video,then the Heart rate(HR)and HRV parameters can be extracted and used for emotional recognition to monitor the emotional changes of the subjects at any time.When using face video to extract HRV parameters,the face area that best expresses heart rate features was found in this paper,that is,the skin part of the face was extracted as the region of interest by using the principle of skin detection,so as to carry out subsequent heart rate feature extraction analysis.In order to solve the problem of the inaccurate extraction of HRV parameters caused by the shaking of the subject's body and head,an abnormal points locating and correction strategy based on the obtained heart rate curve was proposed to explore and correct the irregularity characteristics of the abnormal points in the curve.However,in practical applications,the physical shaking of the subject is an inevitable factor,which leads to inaccurate calculation of HRV parameters.The strategy proposed in this paper uses two threshold ranges of time and heart rate to locate the abnormal points,the mean values of the real-time heart rate were used to correct the position of the abnormal points.Experiments were carried out on 10 volunteers.The pulse waves were simultaneously acquired by video and contact PPG sensor(gold standard).The experimental results show that when applying the strategy in the shake state,the correlation coefficients of HRV parameters(mean,LF,HF,LF / HF)obtained by contactless and contact methods were 0.9615,0.7102,0.7102 and 0.5982,respectively,compared without this strategy,it has increased by 0.0802,0.2785,0.2785 and 0.2677,respectively.The correction strategy proposed in this paper effectively improves the HRV parameters extraction effect based on facial video in the shake state.Emotion recognition using HRV parameters extracted in a non-contact manner was first proposed in this paper,in order to monitor the emotional state of subjects.The pictures in the International Emotional Picture System were used for emotional induction,and the ordinary USB camera was used to collect the facial video data of the volunteers.The HRV parameters were extracted using the strategy proposed in this paper,and the support vector machine(SVM)was used to classify the emotions,the high-low pleasure and high-low arousal are classified separately,and the classification correct rate is higher than 65%.The results show that the HRV parameters collected by the camera are feasible for emotion recognition,which lays a foundation for the application of non-contact physiological signal in emotion recognition. |
