A Real-time Animation Reproduction Based On Active Shape Model Facial Tracking

In everyday life, the facial expression is a very natural method of communication. it is rich in meaning and as an important research topic in the field of human-computer interaction and virtual reality after decades. However, compared with the body movements of sign language, the Facial Expression Applications in business application is not that much.That is because the width of Facial Expression is smaller while contains much detail. This article aim to analyse the project 《facial expressions in real-time using a standard webcam》 worked by professor Yasue Mitsukura who comes from Keio University, and the business software "FaceRig" in the internet.Trying to increase the efficiency in facial expression tracking and improve the stability in gesture recognition algorithm. Second, this article aim to reduce the dependency of algorithm such as good environment and quality device.The main research of this paper include the following aspects:(1)Analyse the papers related to the facial tracking using in 3D animation production. And summarize the difficult problem in this field. The trouble Such as the lower accuracy in first tracking point, the speed and accuracy of tracking processing, and the emergency measures when tracking failure appeared. Finally, propose the improvement and research methods.(2)To solve the facial features must meet the reasonable shape, this article using facial tracking based on ASM to render the 3D animation. Using several major feature points in face to reproduct the posture and facial expression change.(3)The shape of the ASM model of research and reconstruction, integrated into a single low-dimensional projection matrix, so that the algorithm is simple and fast time-consuming processes.(4)Using gradient descent to training ASM texture model, This article accelerate the training speed and quality of training.(5)Improved the initialization module in ASM tracking system, this article using gray projection to locate the human eyes as the the initial point in tracking system, and periodicity setup the initial point to avoid the tracking failure.(6)Aim to improve the detection for new feature point, this article put forward the multiple template matching to detect the new feature point comed from new frame.It improve the accuracy of the tracking algorithm.(7)To solve the lower detection range in posture recognition module, this article using a new postrue recognition algorithm which improve the detection range and accuracy in posture recognition.The innovation of this article is reflected in two aspects. First is the optimization in details in several modules in ASM tracking algorithm. Second is the proposal to using new posture algorithm.