Design And Implementation Of Video-based 3D Face Animation Driven

The techniques of video-driven 3D facial animation is a hotspot in the area of human-computer interaction, which includes the critical information processing technology of face detection, face tracking, facial feature location and extraction, and it is widely used in video conferencing, online games, digital entertainment and so on, so it has a great significance.A video-based 3D facial animation system is designed and implemented in this thesis. The position of facial feature in the image sequences can be located in this system, and the movement of very feature points can be obtained easily. We can use the movement information to control the muscle-based model animation engine and get a relative good result. Our method is a meaningful attempt for realistic 3D facial animation and plays a development role in the application of face recognition technology. Summary full text, the contribution of this paper mainly in the following areas:1. Image pre-processing. In the image pre-processing module, we improve the quality of the image through the image processing, including light compensation, Gaussian smoothing and morphological of gray-scale images.2. Face detection. In the face detection module, we detect the face of the first frame in the video by Viola Jones face detection algorithm, and return the size and position of the face area to initialize the search window of face tracking algorithm in the next frame. To do it like this can also provide a smaller searching range of facial feature location.3. Face tracking. In the face tracking module, we proposed a face tracking algorithm named predicted Meanshift which combining Kalman filter and Meanshift algorithm. This method can track a face with multi-pose or having a temporary occlusion.4. Facial feature tracking. In the facial feature tracking module, an adaptive facial feature location algorithm in video is presented, which using predicted Meanshift and enhanced ASM. The methods in this thesis solve the problem of the inaccuracy and information loss during facial feature tracking, and achieve an automatic, real-time, accurate and robust result in facial feature tracking, so as to provide accurate and effective information for facial animation. In order to keep real time, we depend on CUDA-enabled GPU, which is a novel computing platform where hundreds of on-chip processor cores simultaneously communicate and cooperate to solve time consuming process.5. Translating the feature point information. In this module, we translate the facial feature points obtained from facial feature location algorithm to 3D points in muscle-based animation engine.6. Driving the 3D facial animation. In this module, a video-driven facial animation system is developed based on the 3D feature points.