Research On Video Analysis And Intelligent Diagnosis Of Facial Movement Disorders

With the continuous integration of Artificial Intelligence(AI)and medical fields,AI has been widely used in medical imaging,auxiliary diagnosis and disease prediction.It is of great research significance to apply AI to facial movement disorders recognition which is a common medical condition and realize auxiliary diagnosis.Due to the difficulty of the task itself and the difficult process of dataset construction,there are few researches related to facial movement disorders.In this paper,various computer vision techniques are used to analyze facial movement disorders based on video images,and two recognition algorithms are proposed.The main work of this paper is as follows:Firstly,the related research of micro-expression recognition in fields similar to facial movement disorders recognition is reviewed,and the related work of face alignment is combed.Then the process of constructing the dataset of facial movement disorders is introduced and the movement patterns and characteristics of the two common symptoms are analyzed.Then the samples are preprocessed based on the Convolutional Experts Constrained Local Model and face image segmentation algorithm,and the normalized eye region image sequence is extracted as the research object.Then,a facial movement disorders recognition algorithm based on machine learning is proposed.According to the facial movement rules of patients with different types of facial movement disorders,an integral operator and a difference operator are constructed to calculate the integral average image and motion energy image of image sequences respectively.By analyzing the characteristics of the application scenarios in this paper,the multi-label learning task of facial movement disorders recognition is transformed into two binary classification tasks,the traditional feature extraction method and classification algorithms are used to identify facial movement disorders.The experimental results show that the integral average image and motion energy image can effectively describe different types of facial movement disorders,and achieve good effect.Finally,a facial movement disorders recognition algorithm based on 3-dimensional convolutional neural network is proposed.Aiming at the problem of insufficient complexity of artificial description features and loss of many temporal features,a deep network model is constructed to extract motion features from image sequences.The optical flow method is used to process the original image sequences and superimpose it with the original gray channel as the input of the network.A 3-dimensional convolutional neural network is designed to extract motion information from image sequences,and the long short-term memory is integrated to recognize facial movement disorders.The experimental results show that compared with the traditional machine learning method,the features learned from the deep model are more abundant,which effectively improves the accuracy of facial movement disorders recognition.