Stereoscopic 3D Video Stabilization


Brandon M. Smith
University of Wisconsin - Madison
Li Zhang
University of Wisconsin - Madison
Hailin Jin
Adobe Systems Inc.

ICCV 2011 Demo

Abstract

    Stereo cameras and camera arrays are becoming smaller, cheaper, and more powerful. Even non-professionals will soon have access to handheld devices capable of producing 3D home movies. Unfortunately, handheld cameras often produce very shaky videos despite the best intentions of their users. Camera shake is one of the most distracting aspects of home movies, easily distinguishing them from more professional work. In 3D, camera shake is especially jarring, and potentially nauseating to some.

    To address this problem, we demonstrate a system that produces smooth, stabilized stereoscopic 3D videos from the shaky multi-view output of a handheld camera array. The examples we present depict extremely challenging scenes with little texture and large dynamic targets filmed under severe camera shake. Despite these challenges, our system produces dramatically stabilized 3D results. To the best of our knowledge, this work represents the first software approach for stabilizing shaky stereoscopic 3D videos.

    The results of this work are best viewed on a 3D display. Our demonstration at the conference will therefore include a 3D display and stereoscopic 3D glasses.

 

Previous Work

This demo is an extension of the following work:
Brandon M. Smith, Li Zhang, Hailin Jin, Aseem Agarwala. Light Field Video Stabilization. IEEE International Conference on Computer Vision (ICCV), Sept 29-Oct 2, 2009. Project website: http://www.cs.wisc.edu/~lizhang/projects/lfstable/

Demo Videos

Stereoscopic 3D video [MP4 30.1 MB, WMV 36.3 MB]
This version is best viewed with a 3D video player on a 3D display. We have tested it using the Nvidia 3D Vision Wireless Glasses Kit and an Nvdia 3D Vision-ready laptop and monitor. We assume it can be viewed using other 3D systems as well, but we have not done further tests. On a traditional (2D) video player, the left and right viewpoints will appear simultaneously as side-by-side video streams (including titles and captions).
Red-cyan anaglyph 3D video [MP4 34.0 MB]
This version is best viewed with red-cyan glasses, and is well-suited for viewing on a traditional 2D display.