CONFERENCE (INTERNATIONAL) Optimizing Infinite Homography for Bullet-Time Effect
Ding Chen（Shinichi Higashino）, Ryuuki Sakamoto
The 41st international Conference and Exhibition on Computer Graphics and Interactive Techniques (SIGGRAPH 2014)
September 01, 2014
Bullet-Time camera work is realized when display an object from various aspects by flipping through frames which are taken at same moment by multi cameras surrounding it. Instead of using a robotic pan-tilt head to control the optical axis of the camera, conventional research use tripods to support cameras facing the position of an object approximately, then virtually pan- tilt each camera by applying infinite Homography which, however, causes some blank regions during distorting the original rectangular frame. To eliminate blank regions in the outcome frame, much content is lost as cropping is applied for rectangular displays and as the original aspect ratio is kept. In contrast of conventional research of applying infinite Homography which only corresponds to rotation about optical center with invariable camera intrinsic parameters, our algorithm further optimizes the intrinsic parameters in order to retain as much as content as possible for the outcome frames and achieve the smoothness of Bullet-Time effect. However, if we optimize the intrinsic parameters for infinite Homography separately on each camera, it will result in a very uncomfortable Bullet-Time effect with no continuity in cognitive when changing a frame from one camera to next. Thus our algorithm as the combination of optimizing principal point and focal length of intrinsic parameters achieved the goal of maintaining the smoothness of the effect as well as retaining the original frame content at the greatest extent possible. By applying on several dataset, our algorithm is proved to be remarkably effective compared with conventional research.