Research And Implementation Of Expression Migration Svstem Based On Monocular Camera

Expressions are the carriers for humans to convey emotions and can most intuitively reflect a person's point of view,attitude,and emotion.With the rapid development of entertainment industries such as animated movies and games,people's demand for the expressiveness of virtual characters is also increasing.Among them,expression capture and expression migration based on real performers have become mainstream solutions due to the naturalness and richness of their expressions.At present,most expression capture technologies rely on physical markers or depth cameras,and the equipment requirements are high.In response to this problem,this article studies a facial expression migration system based on a monocular camera,and gives real-time expression capture and video capture based on video images.A model solution for expression synthesis,and based on this,a real-time facial expression migration system is implemented.This paper uses a three-dimensional controllable model to restore three-dimensional facial expressions from two-dimensional images.Based on ResNet,a deep residual 3DMM parameter regression network is established.The network parameters are initialized according to the characteristics of the continuous type of video frames.In order to solve the problem of insufficient 3D image data,a combination of supervised training and unsupervised training is used in this paper.A two-stage loss function is designed to improve the accuracy of expression regression.This paper uses shape fusion algorithm for expression synthesis.Firstly,the control method of the stylized expression-based model is given,and a set of bone hierarchy relationships and skin weight calculation methods are designed.Aiming at the problem that the production of expression-based models requires a lot of manual adjustments,this paper proposes an automatic expression-base synthesis algorithm based on model migration.First,a small number of training expression bases are generated based on Laplacian deformation algorithm and manual fine-tuning.Iterative optimization with fusion weights achieves semantic alignment of all expression bases.The algorithm in this paper reduces the manual workload and improves the fitting effect compared with the traditional expression-based synthesis algorithm.Based on the above three-dimensional expression regression and automatic expression-based synthesis algorithm,combined with face detection,shape fusion and other technologies,a facial expression migration system based on a monocular camera was established.Experiments show that the system can perform high-level facial expressions in videos.The real-time migration of the fit degree verifies the availability of the system.