Research On Dynamic Expression Regresson Model For Real-time Facial Tracking And Animation

The dynamic expression regression model proposed in this paper is essentially tracking the human face in the video,and using the 2D facial information to drive the 3D cartoon model to show the same 3D facial animation.Face recognition and tracking is the focus of artificial intelligence,image processing,pattern recognition and other research fields,and the reconstruction of 3D information based on ordinary 2D video is even more difficult.The traditional 3D shape regression model combines Cascaded Pose Regression Model and Dynamic Expression Model.it uses a single camera to perform accurate and stable facial tracking and animation.However,a long-time calibration step is required for each user before that.Dozens of different head pose and facial expression images of the tracked user are need in the training of the model,which can identify its own characteristics,but can not be effectively applied to the others.In view of the shortage,this paper proposes a non-calibration relatively automatic method.This method can learn a common regression from the public image dataset.When the video object is always the same user,the regression can accurately locate the 2D facial landmarks by any single camera,and then generate the 3D face shape.After that,the 2D facial landmarks can be used to adjust the camera projection matrix and user identities,so as to better match the facial expression of the user.The regression and optimization process is executed alternately,and the feedback loop is effectively created to achieve the adaptation.The entire process converges quickly to obtain more accurate facial tracking and animation,as more and more facial expressions are observed in the video.The related experiments and data analysis figure that:(1)The model achieves a fairly high accuracy and stability under the premise of non-calibration.(2)The model which possesses the universality that the traditional model doesn’t have can be applied to different users with only an one-time training.On the basis of the above research,we design and implement a system of facial tracking and animation based on single camera.The system test shows it can complete the video-based facial tracking,3D cartoon facial animation and so on,and is real-time,accurate and stable.All above contribute to its strong entertainment and consumer application value.