【作者】 李响；

【导师】 耿卫东；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2011， 博士

【摘要】 卡通角色的造型和运动是卡通动画创作的核心。在传统动画制作流程中,动画师通过手绘来设计角色的造型和运动,创作出富有表现力的卡通动画。但传统动画制作流程通常要求动画师绘制大量的动画帧,非常繁琐费时。目前主流的计算机角色动画制作方法一般以几何模型为基础。用户采用基于WIMP(窗口、图标、菜单、指示设备)的交互界面创建模型和制作动画。与手绘相比,这种创作方式复杂且不直观,限制了动画的艺术表现力。本文基于这个出发点,将直观、灵活的手绘创作方式与高效的计算机动画技术相结合,探索了基于手绘图的卡通角色建模与动画生成技术,主要的研究工作包括以下几个方面：1)基于多视点手绘图的卡通角色建模方法。角色的几何建模是计算机角色动画创作的一个关键问题。本文根据卡通角色创作的特点,将角色模型表达为包含网格和骨架的层次化结构,并结合多视点手绘图和领域知识来创建复杂的卡通角色模型。首先结合曲面优化和曲面细分算法,创建与多视点手绘图相匹配的网格模型；然后采用热扩散算法建立网格的骨架子空间模型,以便于角色的运动控制。该方法采取了批处理的建模方式,较为符合传统动画师的创作习惯。2)卡通角色的视点相关变形与绘制方法。传统的三维模型绘制结果通常无法体现手绘动画中常见的视点相关艺术风格。针对这个问题,本文提出了改进的视点相关变形方法。首先结合骨架姿态重建和曲面优化变形算法,使角色模型在每个关键视点与相应的手绘参考图保持一致。给定新的视点,该方法能够生成特定的模型,使其几何特征依赖于相邻关键视点的参考图,并结合轮廓线绘制和纹理映射算法生成卡通风格的画面。该方法通过计算模型与参考图的对应点,减轻了手工标记的负担。3)运动数据驱动的卡通人体动画生成方法。运动数据驱动的人体动画生成技术具有真实、高效的特点。本文基于数据驱动的动画技术改进了视点相关人体动画的生成方法。首先利用手绘图创建一组具有运动状态标注的视点相关模型,然后采用支持向量机技术建立运动数据的标注器。对于具有相似形式的运动数据,该方法可以提取和标注关键帧,并通过启发式泛化算法将视点相关模型的手绘风格应用于输入的运动数据。实验表明,该方法生成的动画能够保留运动数据的典型特征,同时体现手绘图的笔画风格。4)基于手绘图的卡通表情建模方法。在计算机动画领域,如何生成具有抽象、夸张和视点相关效果的手绘表情动画是一项富有挑战性的课题。本文提出了一种体现情感特征和视点相关特征的表情建模方法。首先通过曲线优化变形算法半自动地生成多视点表情集,然后建立基于情感和视点要素的表情空间模型。在此基础上,可以根据新输入的视点和情感状态,较为有效地生成具有特定情感并体现视点相关特征的卡通表情动画。更多还原

【Abstract】 Appearance and motion of cartoon characters play the most substantial roles in cartoon animation. Most traditional cartoon animations are highly expressive, which are created by trained animators who design the appearance and the motion of charaters by drawing. In this process, a large amount of manual work is invested in drawing charcters frame by frame. On the other hand, the conventional computer animation techniques are based on 3D geometry model. User interfaces in modeling and animating follow the WIMP (window, icon, menu, pointer device) paradigm, which are cumbersome and daunting. The work of this thesis is to explore the modeling and animation generation approaches to create cartoon character animation with hand-drawn style, by combining the intuitive sketch-based interfaces and the efficient computer animation techniques. The main research topics of this thesis are as follows:1. Cartoon character modeling based on multi-view hand-drawings. Geometric modeling of character is a key issue in the process of character animation production. This thesis uses a hierarchical structure with mesh and skeleton to represent a character, and presents a character modeling approach based on multi-view drawings and domain knowledge templates. The input drawings are employed to create a complex character mesh matching the multi-view drawings, by combining surface optimization and surface subdivision algorithms. Then the approach constructs the skeleton subspace of the mesh by using heat equilibrium algorithm, making it easy to control the motion of the character. In addition, the batch modeling process of our approach is natural for traditional animators.2. View-dependent deformation and rendering of cartoon character. In traditional animation, a cartoon character often has a view-specific appearance, which cannot be captured with a conventional 3D model. This thesis improves the techniques of view-dependent geometry. For each key viewpoint, the approach deforms the character by skeleton pose reconstruction and surface optimization, yielding a set of models that closely match the hand-drawn reference drawings. Given a novel viewpoint, the approach generates a 3D model that is specific to the new viewpoint, and renders the model to achieve hand-drawn style by silhouette rendering and texture mapping. The approach can compute the correspondence of the base model and the strokes, reducing the amount of manual intervention required by the user in marking corresponding points.3. Cartoon articulated character animation driven by motion data. Motion data is increasingly used to efficiently generate realistic character animation. This thesis presents a data-driven approach for rapidly creating view-dependent character animation. The approach first creates a set of view-dependent models with different motion state annotations from hand-drawn drawings, and then trains annotation classifiers of the motion data by using support vector machines. Given an input motion data with the similar type, the approach extracts and annotates the keyframes, and retargets the hand-drawn style of the view-dependent models onto the input motion data by applying a heuristic generalization technique. The experiment results demonstrate the superiority of this method in generating cartoon character animation with hand-drawn style as well as original motion content.4. Cartoon facial expression modeling based on hand-drawings. How to generate abstract and exaggerated facial expressions with hand-drawn styles is a challenge in the field of computer animation. This thesis presents a hybrid view-and emotion-dependent approach to model cartoon facial expressions. A facial expression space model based on the viewpoint and the emotion is constructed from the multi-view drawings with facial expression, which can be semi-automatically generated by curve deformation. Given a novel viewpoint and an emotion state, the facial expression space model can be used to generate a view-dependent cartoon facial expression with the specified emotion.更多还原