Recovering 3D Geometry from Mutual Projection of Multiple Cameras

点击翻译按钮获取中文摘要

In this research, we investigated visual geometry on the mutual projection of multiple cameras under the case where many cameras exist in the 3D space. As a result, we showed that multi-focal tensors on two-views, three-views and four-views can be compute much more reliably from less corresponding points. We also showed that multiple view geometry for more than four views can be computed reliably by concatenating multi-focal tensors computed from the mutual projection of multiple cameras efficiently.Furthermore, we showed an efficient method for generating free viewpoint images reliably by using the mutual projection of multiple cameras. We also showed a method for navigating multiple robots cooperatively by using the mutual projection of multiple cameras attached on the robots.Since the use of mutual projections of multiple cameras requires the identification of cameras projected in other camera images mutually, we also proposed a method for identifying cameras in images by using the invariance of multiple view geometry.