Archive notice: This article was originally published on July 12, 2010. Links and embedded videos are preserved as part of the historical record.
Keyframes give the computer a start and end point, whose intermediate stages it calculates. According to their own account, Xiaolin Wei and Jinxiang Chai have enriched this technique with several algorithms for Newtonian physics, enabling them to extract 3D data for animation from normal 2D footage. In the abstract of their paper submitted to Siggraph, they describe it as follows:
“This paper presents video-based motion modelling for generating physically realistic human motion from monocular video sequences. We implemented the video-based modelling in a frame-based keyframe-animation framework. The system first examines camera settings, human skeleton size and a small number of 3D poses in order to measure how the motion proceeds in the 2D image.
During motion reconstruction, constraints are imposed on it through Newtonian physics, contact constraints and 2D image measurements in order to measure body posture as well as joint rotations and contact forces – impacts, for example – at the same time. We demonstrate the capabilities and effectiveness of our system by generating realistic human movements from uncalibrated video material, such as sports footage.”
Further Links:
- The paper to read. (PDF)
