Research On Depression Detection Based On Facial Feature Points And Eye Gaze Direction

Depression is a disease characterized by significant and lasting depression and depression.Some patients have self-injury and suicidal behaviors accompanied by delusions,hallucinations and other psychotic symptoms.Depressive stupor may occur in severe cases.At present,the prevalence of depression in society is relatively high,and the common people’s awareness of depression is also insufficient,resulting in fewer patients receiving formal treatment,so it also has the characteristics of higher treatment rate and higher recurrence rate.Based on the data of the 2D feature points of the face and the gaze direction of the human eye in the DAIC-WOZ database,this paper predicts and analyzes whether the subject suffers from depression.By processing the data in the database,extracting characteristic parameters,and substituting them into neural networks and machine learning for training,the applicability of the system is evaluated through parameters such as accuracy and loss.Since facial expressions are affected by factors such as illumination and local facial texture,contour,etc.,and cannot reflect the detailed changes of facial features,this paper uses the deformation of facial muscles to cause changes in the positions of feature points to analyze whether patients suffer from depression..This paper selects the 2D features of the subjects in the DAIC-WOZ database,and converts the 68 feature points of the face into 68*68-dimensional single-channel feature images for feature analysis.Conduct comparative experiments to determine the model framework of the neural network.The final trained network achieves high accuracy on the entire training set and test set,indicating that the network has good performance and generalization ability.Using the 2D feature points of the face,the expression recognition method based on the combination of the feature point pattern and the image pattern can analyze the changes of the facial feature points,which can obtain a more robust depression recognition effect than analyzing the face image.On the other hand,analyzing the gaze data set of human eye gaze direction in the DAICWOZ database,it is found through experiments that when depressed patients are stimulated by negative materials,the deflection angle of their eyes will fluctuate greatly,while non-depressed patients encounter such When stimulated,the amplitude of the change in the deflection angle of the eyes was small.In this paper,the cosine value of the rotation amplitude of the left and right eyes obtained by calculating the fixed frame interval is selected as the transformation of the deflection angle of the subject’s eyes to study.In order to analyze the change of amplitude more clearly,the amplitude spectrogram obtained after Fourier transform is used,and then the extracted amplitude extreme value is used as a feature to analyze the disease condition of the subjects,and then the extracted feature parameters are substituted into the random forest for analysis.Training,through the five-fold cross-validation method,the final accuracy rate of the test set reaches 80.3%,and the experimental results have high accuracy and certain stability.