The Research Of Data-driven Real-time Facial Expression Animation

Data driven real-time facial expression animation could reconstruct model with rich and exquisite facial expression quickly.Scientific workers have researched on facial animation for many years,many novel and unique technical methods were proposed by the excellent research workers,and also a lot of good system was produced.As the hardware performance increased significantly these years,people could capture the facial subtle movement accurately by using advanced motion capture.However,how to realize the reconstruction of the real vivid facial expression model with cheap motion capture in real-time is a very hot topic in the field.The goal of facial expressions reconstruction is to shift the facial expressions of the performer to another neutral facial model,which means to map the motion vector which represents the characteristic points of performer's facial movements to the reconstructed target model through the algorithm processing,using deformation algorithm which has advantage of retaining the geometric feature to reconstruct the realistic and exquisite model.As a large number of applications of special effects on characters in film and animation,how to rebuild the target model with vivid expressions quickly comes to be a popular research direction.In our article,we implement a real-time facial expression reconstruction technology which is based on the data driven.Performers' facial expression data is captured with cheap motion capture device,and the feature points which could represent facial expression movements are chose from the data.Then the movement of feature points is tracked in real-time and a simple grid model is constructed,which covers the whole face.And the feature points and the target model are divided into three parts.The RBF deformation algorithm is adopted to map the motion vector of vertices on source model to the target model,finally blend the three parts of the model to reconstruct a complete target model.In this paper,with a large number of experiments,it's proved that our method could reconstruct high quality and exquisite expression model quickly.In this paper,RBF deformation algorithm and Laplace deformation algorithm are used respectively for real-time facial reconstruction and a new method combing the shortest path and RBF is proposed in this paper.First the facial feature points can be achieved with the motion capture device and some of them are chosen as control points.Second the shortest path between the control points and the other vertices can be figured out with Dijkstra algorithm.The Euclidean distance used in RBF will be replaced by these shortest paths.Finally,the displacement of all points will be calculated to complete the reconstruction of target model.Through the comparative experiment,the method can not only reconstruct the facial expression of source model quickly but also full consider the factor of influence of the control points on the surrounding vertex,which makes the final deformation preserve the geometrical characteristics of the original.