Transcribo el siguiente mensaje del Dr. Luis de la Fraga conteniendo el resumen de Heriberto Cruz Hernández.............................................................................................................From fraga@cs.cinvestav.mx Thu Oct 4 03:55:51 2018Date: Thu, 4 Oct 2018 03:55:49From: Dr. Luis Gerardo de la Fraga <fraga@cs.cinvestav.mx>To: Dr. Guillermo Morales Luna <gmorales@cs.cinvestav.mx>Cc: hcruz@computacion.cs.cinvestav.mxSubject: Resumen de la tesis para seminario de doctoradoEstimado Dr. Guillermo Morales Luna,Le envio el resumen de la tesis de mi estudiante dedoctorado Heriberto Cruz Hernández.Saludos, Luis Gerardo------ Inicio del resumen ------Thesis's title:Joined Solution of Several Computer Vision SubproblemsAbstract:The 3D reconstruction problem consists in obtaining the 3D model of a scene from its projections in multiple images. Inliterature, the task has been treated as the composition of other sub-tasks, i.e., the feature extraction, featurematching, camera calibration, pose estimation, and triangulation. The 3D reconstruction sub-tasks are studied asisolated problems in such a manner that the result of the 3D reconstruction depends on how well the pipeline of tasks issolved. In this work, we study the simultaneous solution of the 3D reconstruction sub-problems. We aim to exploit theinformation of non-contiguous sub-problems by assuming that the information used to solve a sub-task can be used forsolving other sub-tasks. In the first study, we study the simultaneous camera self-calibration, pose estimation and modelretrieving as a non-linear optimization problem. We solve the three sub-problems using as input the point correspondencesin three images. We analyzed the minimal necessary conditions to solve the problem, we proposed a method to includephysical constraints for lines parallelism and orthogonality to solve the camera self-calibration, and we are able toreconstruct models with synthetic but also real datasets.In the second study, we research the joined feature extraction,feature matching and reconstruction of points on a plane. We handled the composed problem through the study of OrderType, a combinatorial invariant feature from the computational geometry field.We show the order type stability during the image generation process, and how with Order Type it is possible to performautomatic identification of point sets on the plane.Furthermore, we propose a method to rank the Order Types regarding robustness to noise and a method to identify thoseOrder Types that are suitable to perform point matching. From our study, we propose a new kind of fiducial markers.Finally, conceptually the two problems are solved simultaneously, using the new markers based in order type toautomatically solve the point matching, and using this result to calibrate a camera and to find the poses on threeimages.---- Fin del resumen ----------------------------------------------------------------------> Dr. Luis Gerardo de la Fraga>> Departamento de Computación> Cinvestav> Av. Instituto Politécnico Nacional No. 2508> 07360 México, D.F.> México> Tel (+ 52) 55 5747 3755. Fax: (+ 52) 55 5747 3757
