Model and Motion Libraries from Video

9972464威廉姆斯，简这个项目研究了从数字化视频中提取一般的三维人体和动物模型及其运动的新方法，以创建一个准确和逼真的虚拟图书馆。逼真的计算机动画人类和动物在计算机图形学和许多其他领域都很重要，包括生物力学、生物学、外科手术规划、运动医学、工程学、教育、培训和人体工程学。自然人体运动模拟在汽车、飞机和工作场所等环境的人体工效学设计中非常重要。虽然对人体类型范围的拟人化研究已经做了很多工作，但人们对特定环境中个人的自然舒适运动范围知之甚少。对人类和动物的运动研究将受益于三维表示，这种表示允许在不同个体和物种的记录运动之间进行定量比较。人类和动物为任何计算机生成的场景增添了真实感和趣味性。录像带已经记录了大量关于人类和许多动物物种自由活动的信息；需要的是从这些记录中提取定量信息的方法。然而，生成复杂铰接结构的真实运动是极其困难的。训练有素的动画师可以使用费力的关键帧技术产生逼真的运动，但结果对科学研究没有用处。在过去的15年里，人们进行了大量的研究，试图使用约束、物理模拟和更高级别的过程规范来部分自动化该过程，取得了一些令人鼓舞的但有限的结果。工作室的运动捕捉技术可以为单个个体提供准确的运动规范，但成本相当高，但推广这些结果的工作才刚刚开始。计算机视觉正在开发自动识别方法，但在这一点上，结果只对简化的情况和环境是可靠的。从在任何一般环境中使用单个摄像机拍摄的视频中提取准确的三维运动的方法将提供必要的通用工具，以将所记录的运动转变为定量有用的形式。这个项目将汇集几个研究领域，以解决提取和表示关节运动的问题。这一过程将结合来自计算机图形学的参数、层次和基于约束的方法，来自生物学和生物力学的生理学见解，以及来自计算机视觉的基于模型的图像合成方法。这一过程将部分自动化，但用户输入将在必要时提供必要的初始条件和更改。结果将是一个准确的人类和动物模型库，伴随着真实的参数化、分层运动描述。加州大学圣克鲁斯分校和其他地方的生物学、生物力学、人类学和视觉等其他研究领域的科学家将参与其中。虚拟图书馆将可用，软件将模块化，以促进与现有系统的整合。

9972464 Wilhelms, JaneThis project investigates new methods of extracting general, three-dimensional human and animal models and their motion from digitized video, in order to create an accurate and realistic virtual library. Realistic, computer-animated humans and animals are important in computer graphics and many other fields, including biomechanics, biology, surgical planning, sports medicine, engineering, education, training, and ergonomics. Natural human movement simulation is important in the ergonomic design of environments such as cars, planes, and work places. While much work has been done in the anthropomorphic study of the range of human body types, less is known about the natural range of comfortable motion among individuals in specific environments. Locomotion studies of humans and animals would profit from three-dimensional representations that allow quantitative comparisons between recorded motion of different individuals and species. Humans and animals lend verisimilitude and interest to any computer-generated scene. Videotape already records massive amounts of information on the free movement of humans and many animal species; what is needed are ways to extract quantitative information from these records. However, generating the realistic motion of complex articulated structures is extremely difficult. Trained animators can produce realistic motion using laborious keyframing techniques, but results are not useful for scientific study. Considerable research has been done in the last 15 years in attempting to use constraints, physical simulation, and higher-level procedural specification to partially automate the process with some encouraging but limited results. Studio motion capture techniques can provide exact motion specification for single individuals, at considerable cost, but work has just begun on generalizing these results. Computer vision is developing automatic recognition approaches, but at this point results are only reliable for simplified cases and environments. Methods of extracting accurate three-dimensional motion from video filmed using a single camera in any general environment will provide the necessary general tool for turning recorded motion into a quantitatively useful form. This project will bring together several research areas to attack the problem of extracting and representing articulated motion. The process will combine parametric, hierarchical, and constraint-based methods from computer graphics, physiological insights from biology and biomechanics, and model-based image-by-synthesis approaches from computer vision. The process will be partially automated, but user input will provide necessary initial conditions and alterations when necessary.The result will be a library of accurate human and animal models accompanied by realistic parameterized, hierarchical motion descriptions. Scientists in other research areas, such as biology, biomechanics, anthropology, and vision, both at the University of California, Santa Cruz, and elsewhere will be involved. The virtual library will be made available, and software will be modular to facilitate integration with existing systems.