CVPR Daily - Tuesday

“ [This] method is the first one that solves the projective structure-from-motion problem directly from a set of fundamental matrices” Yoni thinks that this is a very practical approach . He says that you can just take a set of images which have correspondences and generate a big structure of a scene. They have built several sights that people have captured with their cameras of famous structures in different places around the world (for example: the Église du Dôme in Paris which is showed on the previous page). In terms of next steps, now that the problem has been solved in an uncalibrated setting , Yoni would like to extend the work to a calibrated setting . In the calibrated setting, where the calibration of the cameras is known, he would use essential matrices instead of fundamental matrices and then generate a calibrated structure which means that the angles of the scene are defined correctly. In the projective set- up, the angles are not unique and there is some projective ambiguity, but in the calibrated setting, the angles are well defined. If anyone wants to use this method, the code is available here . To find out more about this work, come along to the poster session [79] today (Tuesday) from 15:20-18:00. “ This is a very practical approach!” 17 DAILY CVPR Tuesday GPSfM