Acquisition and Modeling of Non-Rigid Shape and Deformation

非刚性形状和变形的获取和建模获取真实世界物体的三维数字复制品可以实现数字存档、虚拟修复、仿真、医学、教育、逆向工程和许多其他应用。最近，米开朗基罗的雕塑、佛罗伦萨的圣母像和开罗埃及博物馆的文物数字化工作取得了成功，这表明扫描静态、刚性物体的技术挑战在很大程度上已经得到了解决。然而，同样的扫描技术对于许多需要对非刚性、变形物体（包括人类、动物和环境）进行数字化的应用来说，效率要低得多。例如，当今最先进的扫描仪要求病人保持一个姿势几秒钟不动，即使婴儿、小孩和残疾人可能很难在规定的时间内保持静止。更根本的是，当前扫描技术的主要缺点是，对可变形物体的单一静态扫描不能很好地反映其形状，因为人类和动物的形状随时都可能发生变化。本研究解决了获取移动三维形状的算法挑战。使用高速深度相机获取实时距离图像允许物体在整个扫描过程中自由变化，并在此过程中显示其整个形状，包括通常被遮挡的区域。然后对所获取的距离图像进行自动处理，以计算附近帧中数据点之间的时间对应关系；反转这些数据点的运动；并通过聚合每个时间框架的部分信息来重建整体的运动三维形状。这些算法简化了非刚性形状的获取，并实现了新的应用，例如人类和动物的运动发展研究以及教育和培训的内容创建。

Acquisition and Modeling of Non-Rigid Shape and DeformationThe acquisition of three-dimensional digital replicas of real-word objects enables digital archiving, virtual restoration, simulation, medicine, education, reverse engineering, and many other applications. The recent success of the efforts to digitize sculptures by Michelangelo, the Florentine Pieta, and artifacts in Cairo's Egyptian Museum demonstrate that the technical challenges involved in scanning static, rigid objects have been largely solved. However, the same scanning technology is far less effective for numerous applications requiring digitization of non-rigid, deforming objects, including humans, animals, and the environment. For example, the state-of-the-art scanners today require human patients to remain motionless in a single pose for several seconds even though infants, small children, and people with disabilities may find it difficult to remain still for the required time periods. More fundamentally, the key drawback of present-day scanning technology is that a single, static scan of a deformable object is a poor representation of its shape because human and animal shapes can easily change at any time instant.This research tackles the algorithmic challenges in acquiring moving three-dimensional shapes. Using high-speed depth cameras to acquire real-time range images allows the object to change freely throughout the scanning process and in the process reveal its entire shape, including the usually occluded regions. The acquired range images are then automatically processed to compute temporal correspondence between data points in nearby frames; to reverse the motion of these data points; and to reconstruct the moving three-dimensional shape in its entirety by aggregating partial information from each time frame. These algorithms simplify acquisition of non-rigid shapes and enable new applications such as studies of locomotion development in humans and animals and content creation for education and training.