Un modèle géométrique multi-vues des taches spéculaires basé sur les quadriques avec application en réalité augmentée

Abstract : Augmented Reality (AR) consists in inserting virtual elements in a real scene, observed through a screen or a projection system on the scene or the object of interest. The augmented reality systems can take different forms to obtain a balance between three criteria: precision, latency and robustness. It is possible to identify three main components to these systems: localization, reconstruction and display. The contributions of this thesis focus essentially on the display and more particularly the rendering of augmented reality applications. Contrary to the recent advances in the field of localization and reconstruction, the insertion of virtual elements in a plausible and aesthetic way remains a complicated problematic, ill-posed and not adapted to a real-time context. Indeed, this insertion requires a good understanding of the lighting conditions of the scene. The lighting conditions of the scene can be divided in several categories. First, we can model the environment to describe the interaction between the incident and reflected light pour each 3D point of a surface. Secondly, it is also possible to explicitly the environment by computing the position of the light sources, their type (desktop lamps, fluorescent lamp, light bulb, . . . ), their intensities and their colors. Finally, to insert a virtual object in a coherent and realistic way, it is essential to have the knowledge of the surface’s geometry, its chemical composition (material) and its color. For all of these aspects, the reconstruction of the illumination is difficult because it is really complex to isolate the illumination without prior knowledge of the geometry, material of the scene and the camera pose observing the scene. In general, on a surface, a light source leaves several traces such as shadows, created from the occultation of light rays by an object, and the specularities (or specular reflections) which are created by the partial or total reflection of the light. These specularities are often described as very high intensity elements in the image. Although these specularities are often considered as outliers for applications such as camera localization, reconstruction or segmentation, these elements give crucial information on the position and color of the light source but also on the surface’s geometry and the material’s reflectance where these specularities appear. To address the light modeling problem, we focused, in this thesis, on the study of specularities and on every information that they can provide for the understanding of the scene. More specifically, we know that a specularity is defined as the reflection of the light source on a shiny surface. From this statement, we have explored the possibility to consider the specularity as the image created from the projection of a 3D object in space.We started from the simple but little studied in the literature observation that specularities present an elliptic shape when they appear on a planar surface. From this hypothesis, can we consider the existence of a 3D object fixed in space such as its perspective projection in the image fit the shape of the specularity ? We know that an ellipsoid projected perspectivally gives an ellipse. Considering the specularity as a geometric phenomenon presents various advantages. First, the reconstruction of a 3D object and more specifically of an ellipsoid, has been the subject to many publications in the state of the art. Secondly, this modeling allows a great flexibility on the tracking of the state of the specularity and more specifically the light source. Indeed, if the light is turning off, it is easy to visualize in the image if the specularity disappears if we know the contour (and reciprocally of the light is turning on again). (...)
Document type :
Complete list of metadatas

Cited literature [60 references]  Display  Hide  Download

Contributor : Abes Star <>
Submitted on : Thursday, May 23, 2019 - 9:31:00 AM
Last modification on : Friday, May 24, 2019 - 1:39:09 AM


Version validated by the jury (STAR)


  • HAL Id : tel-02137435, version 1


Alexandre Morgand. Un modèle géométrique multi-vues des taches spéculaires basé sur les quadriques avec application en réalité augmentée. Imagerie médicale. Université Clermont Auvergne, 2018. Français. ⟨NNT : 2018CLFAC078⟩. ⟨tel-02137435⟩



Record views


Files downloads