Abstract
We have devised multi-viewpoint robotic cameras for capturing multi-viewpoint images of subjects moving in 3D space with panning shots and zoom-in shots. The camera operator instructs a robotic camera to fix its gaze point at an arbitrary position in 3D space and other robotic cameras automatically follow it as their gaze point. The operator can then capture multi-viewpoint images by fixing the gaze point on the subject position. Bullet time and integral 3D imagery are generated from the images shot by the cameras. Bullet time is generated by using a virtual camera control with a projective transformation to make the viewpoints change smoothly. Integral 3D images are generated by converting a 3D model of a subject created using the depth estimation method into elemental images. In experiments, robotic cameras were able to capture multi-viewpoint images of a sports player moving in 3D space and bullet time and integral 3D imagery were generated from these images. Furthermore, these robotic cameras were used to generate bullet time in a live broadcast of a figure skating competition. In particular, bullet time was used to show the skaters' jumps in instant replays right after each performance.
