CCF: Capturing and Animating the Human Hand: Robust Recovery of Hand-Object Interactions

CCF：捕捉和动画人手：手物体交互的稳健恢复Nancy Pollard摘要这项研究解决了准确捕捉人手运动的问题。由于各种原因，捕捉人类手的运动被证明是异常困难的，包括手中关节的复杂性、人与人之间手解剖的差异、运动范围小、与运动捕捉技术相关的困难以及操纵物体时复杂的接触条件。然而，手对于沟通、关心他人和自己，以及使用工具改变我们周围的世界都至关重要。这项研究涵盖了一套实用技术，使使用当今的硬件准确、特定于对象地捕捉人类的手成为可能。这些技术与相应的手部运动数据库一起，将使在使用捕获的数据以高细节和准确性执行各种任务时研究手部的工作成为可能。这项工作应该有助于康复，以衡量在改善活动范围和完成日常活动方面的进展。这对于探索机械手的潜在设计应该是有用的。这对于探索灵巧性本身的基础也大有裨益。第一个是用于从运动捕捉数据中自动提取特定于受试者的骨骼模型的稳健算法。然后我们观察到，标记协议对结果影响很大。因此，这项研究的第二个小主题是为准确的数据捕获获得有充分依据的标记协议建议。接下来，我们观察到，清理后的运动捕捉数据没有捕捉到对理解抓握非常重要的接触条件。因此，这项研究的第三个子课题是开发优化算法，以获得观察到的运动的物理上可信的表示，即使在手的可变形组织和被操纵对象之间具有复杂的、变化的接触的情况下也是如此。

CCF: Capturing and Animating the Human Hand: Robust Recovery of Hand-Object InteractionsNancy PollardAbstractThis research addresses the problem of accurately capturing motion of the human hand. Capturing human hand motion has proven exceptionally difficult for a variety of reasons, including the complexity of the joints in the hand, variation in hand anatomy from person to person, small ranges of motion, difficulties related to motion capture technology, and complex contact conditions while manipulating objects. However, the hand is critically important for communication, caring for others and ourselves, and using tools to alter the world around us. This research covers a suite of practical techniques needed to make accurate, subject-specific capture of the human hand possible using today's hardware. These techniques, along with the corresponding hand motion database, will make it possible to study the workings of the hand when performing all manner of tasks with captured data at a high level of detail and accuracy. This work should be useful in rehabilitation for measuring progress in improving range of motion and in accomplishing everyday activities. It should be useful for exploring potential designs for robot hands. And it should also be of great use for exploring the basis of dexterity itself.This research involves three specific subtopics. The first is robust algorithms for automatic extraction of subject specific skeletal models from motion capture data. We then observe that marker protocol affects the results a great deal. Thus, the second subtopic of this research is obtaining well founded recommendations for marker protocols for accurate data capture. Next we observe that cleaned motion capture data does not capture the contact conditions that are so important to understanding grasping. Thus, the third subtopic of this research is developing optimization algorithms to obtain physically plausible representations of observed motion, even in situations with complex, changing contacts between deformable tissues of the hand and a manipulated object.