Por favor, use este identificador para citar o enlazar este ítem:
http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/184| Registration of a three-dimensional model of a human head over a scene using color and depth information | |
| MIGUEL DE JESUS OSORIO RAMOS | |
| LEOPOLDO ALTAMIRANO ROBLES | |
| Acceso Abierto | |
| Atribución-NoComercial-SinDerivadas | |
| Augmented reality Computer science Computer vision Computer applications Medical information processing | |
| The objective of Augmented Reality (AR) is to enhance a user’s perception of an environment in real life by providing useful additional information. To achieve this objective, virtual objects must be correctly positioned in a real environment to improve the sense of coexistence between real and virtual objects. In medicine, AR has been well received as it has shown great potential when aiding in diagnosis, surgical planning or surgical procedures because specific regions can be precisely targeted. The observation of spatial relations between the virtual objects and the environment are essential in minimally invasive procedures. Regarding head procedures, a precise registration is of utmost importance. Even though there has been research in this area, further investigation can lead to invaluable discoveries. When using low-cost computer vision systems, the aligning task becomes difficult because of excess noise. Using the scene’s depth information obtained from low-cost range sensors has shown good results; however, these results are still far from being dependable for the medical field. This thesis addresses the issue of registering a virtual model with a human head where images are acquired from two different sources. The first one is an X-Ray Computed Tomography (CT) scan of the head which is used to create a three-dimensional (3D) virtual model. The second is the augmented scene, obtained through a low-cost range sensor taking advantage of depth information. The proposed method aims at reducing registration errors between the 3D virtual model of the head and the one in the real world scene. Our method involves a series of segmentation processes to bound the region of interest, a preliminary head pose estimation which serves as a basis for the next step, and a final pose refinement through the Iterative Closest Point algorithm in order to reach the goal of precise alignment of a human head within a virtual environment. We tested the registration results of the proposed method against the registration of other approaches found in literature addressing the same problem and using low-cost computer vision systems. For evaluation, we built a new database which was manually labeled with the transformation values to align the model with a scene. The reliability of the database was evaluated using an inter-rater reliability test in order to check the agreement of the raters. | |
| Instituto Nacional de Astrofísica, Óptica y Electrónica | |
| 2014-08 | |
| Tesis de maestría | |
| Inglés | |
| Estudiantes Investigadores Público en general | |
| Osorio-Ramos M.J. | |
| CIENCIA DE LOS ORDENADORES | |
| Versión aceptada | |
| acceptedVersion - Versión aceptada | |
| Aparece en las colecciones: | Maestría en Ciencias Computacionales |
Cargar archivos:
| Fichero | Descripción | Tamaño | Formato | |
|---|---|---|---|---|
| OsorioRMJ.pdf | 26.46 MB | Adobe PDF | Visualizar/Abrir |