CAREER: Reusable, Realistic Motion Libraries for Computer Animation

动画是一种非常强大的表达媒介。用静态图像或文字来表达的复杂甚至不可能表达的想法可以很容易地用动画序列来传达。不幸的是，即使通过动画传达的概念很简单，制作动画的过程却异常困难。的确，这是电脑动画占据主导地位的原因之一——它有望实现自动化，从而大大简化制作过程。可悲的是，将想法转化为电脑动画的过程仍然很不简单。例如，动画电影《玩具总动员2》需要多达200名高技能的计算机和动画专家花费三年多的时间进行极其艰苦的工作。我的职业目标是开发概念和算法，使轻松，快速地开发逼真的动画，有效地传达创作者的目的。这项研究希望为构建比现在更强大的动画工具创造理论和实验基础。这些工具将大大减少熟练的动画师所需的制作时间，更重要的是，使普通计算机用户能够通过动画表达他们的想法。这些工具还将帮助教育者更有效地教学，并使我们每个人都成为个人故事和虚构故事的创造者和导演。只有当动画制作变得更容易时，动画才能像今天的图像和文本一样在网络上普及。此外，创造逼真的人类动作的直观方法将进一步使远程呈现应用、新拟人化人机界面以及故事片和视频游戏中的逼真数字演员的创造成为可能。除非1)动画工具只需要很少或不需要技巧，2)动画师的创造力不受动画工具的限制，3)每个拥有电脑和故事的人都可以使用动画系统，否则计算机动画将无法充分发挥其潜力。我希望为实现这些目标作出重大贡献。这项研究的核心是观察到动画师的创造力将不会完全实现，直到它变得更容易产生逼真的运动序列。这个问题极具挑战性，因为运动的基本物理模型很难创建和控制。我相信解决方案在于创建高度灵活的逼真运动库，以及直观和广泛地修改它们的工具。库的强大源于重用已经存在的高质量运动的能力，减少了对动画技能的需求。在这个范例中，动画的过程变成了选择特定的运动库，并修改一组运动属性，将原始运动转换为最终动画。在我的研究中，我建议开发一种方法来创建和使用足够灵活的现实运动库，以用于广泛的应用。1995年，我提出使用运动变换作为创建动画的新方法——一种与从零开始创建动画的传统方法根本不同的方法。这种转换方法对于修改从真实世界的演员那里捕获的真实运动数据特别有用。不幸的是，在转型过程中，许多现实主义倾向于丢失。最近，我发表了一种新颖的转换方法，该方法表明动画可以直观地转换为广泛的新序列，而不会违反运动的基本动态特性。通过保持动画角色的动态和生物力学特性的模型来保持现实性。在未来，我将进一步开发数学模型，可以与现实世界的数据相结合，以创建可重用的运动库。除了保持运动的真实性，对有效的运动库的最重要的要求是库可以调整的灵活性，以满足动画师的需要。我计划通过将角色的运动分解为基本组件和风格组件来实现这种灵活性，从而允许我们独立地将这些组件转移到新角色中。例如，动画师可以通过从捕获的人类跑步中提取的runningmotion库开始，生成孩子欢快的跑步序列。这个动作可以通过应用一个快乐的、充满活力的动作风格来修改，然后转移到一个小孩的动画角色上。这种基本运动、风格和表演角色的解耦提供了一个非常强大和灵活的动画范例，可以从一个非常小的运动数据集产生广泛的动画。本提案的教育部分的主要目标是展示先进的动画工具在教育和讲故事中的效用。我特别感兴趣的是向西雅图的高中生介绍这些新的动画工具，以验证我们的研究确实为每个讲故事的人提供了一种可访问的表达媒介。我还计划开设一个项目导向的动画课程，面向从事非动画研究的本科生和研究生，以及艺术和音乐方面的学生。学生们将一起使用我们的工具来创作一个艺术短片项目。这种协同的努力将使艺术家接触到最新的动画技术，并为参与研究的学生提供直接的用户反馈。此外，制作的动画库和动画工具将通过网络和cd免费分发，希望它们能帮助动画成为跨越教育、经济和文化界限的有效交流媒介。

Animation is a tremendously powerful medium of expression.Ideas that would be complex or even impossible to expresswith still images or words can easily be conveyed usinganimated sequences. Unfortunately, even when the concept tobe conveyed via animation is simple, the process ofproducing the animation is inordinately difficult. Indeed,this is one of the reasons computer animation has become sodominant - it holds the promise of automating, and thussignificantly simplifying, the production process. Sadly,the process of transferring ideas into computer animationsis still far from simple. For example, the animated featurefilm Toy Story 2 required extremely hard work by as many as200 highly skilled computer and animation experts for morethan three years.My career goal is to develop the concepts and algorithmsthat will enable effortless, rapid development of realisticanimations that effectively convey the creator's purpose.This research hopes to create the theoretical andexperimental foundation for building much more powerfulanimation tools than exist today. Such tools will sharplyreduce the production time needed by skilled animators and,more importantly, enable average computer users to expresstheir ideas through animation. Such tools will also helpeducators teach more effectively and allow each one of us tobecome creators and directors of personal and fictionalstories. Only when animation production is easier willanimation be as commonplace on the web as images and textare today. Moreover, intuitive methods for creatingrealistic human motion will further enable the creation ofhuman avatars in tele-presence applications, newanthropomorphic human-computer interfaces, and realisticdigital actors in feature films and video games. Computeranimation will not achieve its full potential until 1)animation tools require little or no skill, 2) an animator'screativity is not stifled by the limitations of theanimation tools, and 3) the animation systems are availableto every person with a computer and a story to tell. I hopeto significantly contribute towards these goals.At the heart of this research is the observation that theanimator's creativity will not be fully realized until itbecomes easier to produce realistic motion sequences. Thisproblem is tremendously challenging because the underlyingphysical models of motion are difficult to create andcontrol. I believe that the solution lies in the creation ofhighly-flexible realistic motion libraries, together withtools to modify them intuitively and extensively. The powerof libraries stems from the ability to reuse alreadyexisting high-quality motion, reducing the need foranimation skills. In this paradigm, the process of animationturns into selecting the specific motion library, andmodifying a set of motion properties that transform theoriginal motion into a final animation.In my research, I propose to develop a methodology forcreating and using realistic motion libraries flexibleenough to be used in a wide range of applications. In 1995,I proposed the use of motion transformation as new way tocreate animations - a method fundamentally different fromthe traditional way of creating animations from scratch.This transformation approach is particularly useful formodifying realistic motion data captured from real-worldactors. Unfortunately, during the transformation process,much of the realism tends to be lost. Recently, I publisheda novel transformation approach which demonstrates thatanimations can be intuitively transformed into a wide rangeof new sequences without violating the fundamental dynamicproperties of motion. The realism is preserved bymaintaining a model of the dynamic and biomechanicproperties of the animated character.In the future, I will further develop mathematical modelsthat can be combined with real-world data to create reusablemotion libraries. Aside from preserving the realism ofmotion, the most important requirement for effective motionlibraries is the flexibility with which the libraries can beadjusted to meet the needs of an animator. I plan to achievethis flexibility by decomposing a character's motion into afundamental component and a style component in a waythat allows us to independently transfer these components tonew characters. For example, the animator can produce achild's cheerful run sequence by starting with a runningmotion library extracted from the captured human run. Thismotion can be modified by applying a happy, exuberant motionstyle, which is then transferred to an animated character ofa small child. Such decoupling of the fundamental motion,style, and performing character provides an extremelypowerful and flexible animation paradigm that can produce awide range of animations from a very small motion dataset.The main goal of the educational component of this proposalis to demonstrate the utility of advanced animation tools ineducation and story-telling. I am particularly interested inintroducing these new animation tools to Seattle high-schoolstudents to verify that our research does indeed empowerevery story-teller with an accessible medium of expression.I am also planning a project-oriented animation courseintended for undergraduate and graduate students involved inanimation research, as well as in art and music. Together,the students will work on using our tools to create anartistic short film project. This synergistic effort willexpose the artists to the latest animation technology andprovide direct user feedback to students involved inresearch. In addition, the created motion libraries as wellas the animation tools, will be freely distributed throughthe web and CDs with the hope that they will help animationbecome an effective communication medium across educational,economic and cultural boundaries.