Research On Pain Expression Recognition Algorithm Based On Dynamic Bayesian Network

Pain is considered one of the important vital signs,and for patients with speech impairment,timely use of patients’ facial expressions to determine pain or provide information about the patient’s pain level can help healthcare workers better treat patients,so it is especially important to carry out accurate recognition and analysis of facial pain expressions.In practical applications,there are fewer sample data of pain expressions available for training and testing,and in addition,there are various uncertainties in the process of obtaining pain expressions,such as uneven lighting,face posture changes,and occlusion.These factors can cause inaccurate or missing facial features,which in turn affect the results of facial pain level assessment.Bayesian Network(BN)is a method that combines probability theory and graph theory,and has significant advantages in building and reasoning about uncertainty problems.Therefore,in this paper,a Dynamic Bayesian Network(DBN)based pain expression recognition method is designed in order to improve the accuracy of pain recognition of facial expressions and considering the temporal correlation between pain intensities,as follows:(1)In order to quickly and accurately obtain the pain expression feature samples required for pain expression recognition,this paper investigates the process of acquiring facial action units(AU)for pain expressions of human faces.First,the Viola-Jones face detection algorithm is used to detect the facial images;then the constrained local neural domain model is used to extract the key feature points from the facial expression images;finally,the support vector machine is used to obtain the pain-related AU feature sample set.(2)In order to reduce the impact of unbalanced distribution of existing pain expression data on the accuracy of pain expression recognition,this paper combines random oversampling and random undersampling to balance the initial AU feature sample set under the condition of maintaining sample authenticity.(3)In order to improve the accuracy of pain expression recognition,this paper combines traditional BN with temporal information to construct a DBN model for pain expression recognition that can effectively exploit the temporal correlation between data.The structure determines the number of network nodes and the directed edges required for the DBN structural model of pain expression recognition by analyzing the relationship between facial AU and pain expressions and combining expert experience.(4)For the problem of low accuracy of traditional BN parameter learning algorithm under small sample conditions,a BN parameter learning algorithm(Two-level Estimation Interval,TLEI)based on the dual estimation interval is proposed.Firstly,the qualitative size relationship between parameters is used to obtain a priori parameter distribution intervals;secondly,the Beta distribution is used to fit the parameter distribution and generate random numbers obeying the Beta distribution,i.e.,virtual parameters;then the temporary sample sets obtained based on the nonparametric Bootstrap method are fused with the parameter samples generated based on the parametric Bootstrap method to satisfy the constraints with variable weights to obtain new The maximum and minimum values of each parameter are selected to construct the parameter estimation interval;finally,the dummy parameters falling into the interval are counted,and the hyperparameters in the maximum a posteriori probability algorithm are corrected using the statistical results,and then the BN parameters are calculated.After experimental validation on the standard Weather network and Asia network,the experimental results show that the parameter learning accuracy of TLEI algorithm is better compared with that of maximum likelihood estimation algorithm,maximum a posteriori probability algorithm and qualitative maximum a posteriori probability algorithm on small data sets.(5)To address the problem that sparse pain sample data and uncertainties in recognition can affect pain expression recognition in practical applications,this paper designs a pain expression recognition method based on DBN and TLEI algorithm,and conducts experimental validation on UNBC-McMaster shoulder pain expression database.Firstly,the pain-related AU sample dataset in the image is acquired and balanced;secondly,the DBN structure for pain level evaluation is constructed;then the TLEI algorithm is used to learn the parameters of this structure;finally,the 1.5 piece joint tree algorithm is used to complete the recognition of pain expressions.The experimental results show that the pain expression recognition method based on DBN and TLEI algorithm has a high recognition accuracy.