The increasing availability of video capture devices makes it easy to capture a video. However, creating a video with high-quality camera motion is still difficult. It requires equipment, effort, and expertise. The camera motion in most videos taken by common users looks amateurish. Even some professional videographers sometimes complain about the difficulty of achieving good camera motion. Moreover, the emerging ways that people view videos further complicate the problem. Videos need to be adapted to different applications than originally intended. The transformation will damage the original good camera motion. Existing methods to post-process videos to improve their camera motion or adapt them to new applications are limited. They either ignore cinematic rules or lack the capability to assist people in achieving an idealized camera motion. This dissertation shows how to develop technologies that create high-quality camera motion for videos by borrowing professional expertise and exploiting visual plausibility. Accordingly, this dissertation presents three technologies that assist people in either easily creating a good apparent motion of a video or preserving it. The first is a 3D approach to transforming a video so that it has an idealized camera path. The key technical contribution is the design and use of the content-preserving warping that enables this 3D approach to work well with scene dynamics. The second is a subspace video stabilization approach that enables aggressive smoothing. The key technical contribution is the proposal of estimating eigen-trajectory as a suitable motion representation for video stabilization that is easy to compute and effective for the stabilization. This approach is robust, effective, efficient and streamable, and practical enough for an end-user tool. The third is a video retargeting method to adapt edited videos to heterogenous displays. This method extends and automats pan and scan and enforces cinematic plausibility during video adaptation.

@PhdThesis{Liu10,
  author       = "Liu, Feng",
  title        = "Technologies for Creating Good Camera Motion",
  school       = "University of Wisconsin Madison",
  year         = "2010",
  projecturl   = "http://web.cecs.pdx.edu/~fliu/thesis/",
  url          = "http://graphics.cs.wisc.edu/Papers/2010/Liu10"
}