Data-Driven Animation of Skin Deformations

数据驱动的皮肤变形动画光学运动捕捉已经非常成功地用于为娱乐应用程序创建引人注目的人体动画；然而，它只提供了一个非常简化的版本，如果我们要查看一个实际执行这些动作的人。这些数据包含骨骼的运动，但忽略了肌肉膨胀和肌肉抖动等视觉上的重要效果。目前的技术水平使用40-60个标记来近似15-22个节段的刚体运动。尽可能地将标记放置在关节轴和骨骼地标上，以便可以更容易地使用它们来找到理想骨骼的运动。研究人员正在采取一种不同的运动捕捉方法，并使用一组非常大的标记(约350-500个)放置在身体的肌肉和肉质部分，不仅捕捉骨骼的运动，还捕捉皮肤表面的运动。皮肤和肌肉变形模型将在计算机动画、视频游戏、外科手术规划、类人机器人和生物工程等领域得到长期应用。研究人员正在使用捕获的数据来构建由关节角度、速度和扭矩参数化的人体运动变形的动态模型。这些模型被用来将捕获的数据概括为具有相似身体类型的新运动和新受试者。研究人员通过将这种密集标记集的三维轨迹应用于受试者特定的多边形模型，已经能够准确地重建身体表面的运动。他们正在探索一些不同的方法来推广这些数据，从简单的统计动态模型到从数据中识别参数的相对精确的解剖模型。

Data-driven Animation of Skin DeformationsOptical motion capture has been used very successfully to create compelling human animations for entertainment applications; however, it provides only a much simplified version of what we would see if we were to view a person actually performing those actions. The data contain the motion of the skeleton but miss such visually important effects as the bulging of muscles and the jiggling of flesh. The current state of the art uses 40-60 markers to approximate the rigid body motion of15-22 segments. To the extent possible, the markers are placed on joint axes and bony landmarks so that they can more easily be used to find the motion of an idealized skeleton. The investigators are taking a different approach to motion capture and use a very large set of markers (approximately 350-500) placed on the muscular and fleshy parts of the body to capture not only the motion of the skeleton but also the motion of the surface of the skin. Models of skin and muscle deformations will have long term applications in such fields as computer animation, video games, surgical planning, humanoid robotics, and bioengineering.The investigators are using captured data to construct dynamic models of the deformations seen in human motion parameterized by joint angle, velocity, and torque. Those models are used to generalize captured data to new motions and new subjects with a similar body type.The investigators have been able to accurately reconstruct the motion of the surface of the body by applying the three-dimensional trajectories for this dense marker set to a subject-specific polygonal model. They are exploring a number of different approaches to generalize this data ranging from simple statistical dynamic models to relatively accurate anatomical models with parameters identified from the data.