Generating Free Viewpoint Images from Mutual Projection of Cameras

In augmented reality, accurate geometric adjustment of real scene and virtual 3D models is important. In this paper, we propose a new method for generating arbitrary views of 3D motion events accurately by using the mutual projections between user’s cameras and cameras around the user. In particular, we show that the trifocal tensors computed from the mutual camera projections can be used efficiently for generating accurate user’s views of 3D motion events from multiple camera images. We also show a method for identifying cameras projected in other cameras by using the invariance in multiple view geometry. The proposed method is implemented and tested in the real scene.