Character animation is a challenging research topic in computer graphics, and has a very wide range of applications, such as film and television production, video games, virtual reality and military simulation. Its goal is to create natural and realistic characters in virtual environment, focusing on model motion control and interaction between characters and the environment. Motion capture is an efficient method to generate motion data that is of high fidelity and rich detail. However, it can only be applied in certain circumstances, and prone to self-collision and collision between models. Although there has been a lot of valuable and practical study on character animation both at home and abroad, however, with the increasing requirements for efficient and realistic movements, it is still a challenge to perform high quality simulation in an interactive virtual environment. Focusing on several issues when motion capture data applied to characters in virtual environments, the paper implements and experiments a group of novel algorithms in three perspectives: human skin deformation, continuous collision detection for deformable objects and physics-based collision response. Main contributions of this thesis are as follows: For the reality problem of model deformation, a parameterized human body model and its skin deformation method was proposed. According to the principles and methods of anthropometry, a group of body parameters (such as arm length, waist, etc.) was determined, and a region partition mechanism was introduced to divide the model into rigid and soft regions. The free-form deformation method based on SCODEF was extended, so we can stretch the rigid region by using body parameters and region constraints. In the mean time, we use dual quaternion linear blending method to twist the soft region. Our method can easily get stretch and twist deformation without artifacts like candy-wrapper and collapse. In addition, the skin weight adjusted by automatic rigging technique further improved the deformation quality. Aiming at self-collision of character models and collision between characters and objects, a continuous collision detection method based on octree was proposed. Unlike traditional methods that divide the whole space, we partition static and dynamic objects respectively. For static objects, a top-down approach was used to construct their octree and normal cone was introduced to prevent excessive division; while for dynamic objects, they ar...
修改评论