Imagerie RTI adaptative : Acquisition, Automatisation et mosaïcage.

Adaptive Reflectance Transformation Imaging : Acquisition, Automation and Stitching.

• Auteur / Autrice : Luxman Ramamoorthy
• Direction : Alamin Mansouri, Franck Marzani, Gaëtan Le Goïc
• Type : Thèse de doctorat
• Discipline(s) : Instrumentation et informatique de l'image
• Date : Soutenance le 21/09/2023
• Etablissement(s) : Bourgogne Franche-Comté
• Ecole(s) doctorale(s) : École doctorale Sciences pour l'ingénieur et microtechniques (Besançon ; 1991-....)
• Partenaire(s) de recherche : Laboratoire : Imagerie et Vision Artificielle (ImVia) (Dijon)
• Partenaire(s) de recherche : Etablissement de préparation : Université de Bourgogne (1970-....)
• Jury : Président / Présidente : Christian Daul
• Jury : Examinateurs / Examinatrices : Jon Yngve Hardeberg
• Jury : Rapporteurs / Rapporteuses : Robert Sitnik, Guillaume Caron

Mots clés

Mots clés contrôlés

Robotique

Infographie

Vision artificielle (robotique)

Mots clés libres

Reflectance Transformation Imaging

Apparence

Vision pour la Robotique

Acquisition adaptative

Vision artificielle

Reflectance Transformation Imaging

Appearance

Vision for robotics

Adaptive acquisition

Computer Vision

Résumé

Le Reflectance Transformation Imaging (RTI) est une technique d'imagerie mesurant la réflectance angulaire locale des surfaces en variant l'angle d'éclairage. Utilisée pour étudier le patrimoine culturel, elle révèle des attributs de surface invisibles à l'œil nu ou à la lumière ambiante. Le RTI capture des images d'un objet éclairé sous différents angles sur un hémisphère ou un dôme. Les images sont combinées pour créer une seule image haute résolution codant les propriétés de réflectance de la surface. L'image RTI résultante est visualisable et ajustable virtuellement pour mettre en évidence des caractéristiques de surface. L'acquisition, la modélisation et la visualisation de la réflectance des surfaces complexes sont un domaine de recherche actif. Pour cela, nous avons développé des méthodes d'acquisition RTI adaptative et d'automatisation. Nous avons estimé la configuration d'éclairage optimale et amélioré la résolution des données RTI par assemblage. Ces avancées visent à améliorer l'efficacité de la RTI et l'imagerie numérique du patrimoine culturel. Les acquisitions RTI sont généralement manuelles ou utilisent des dômes RTI, mais nous avons exploré l'utilisation de bras robotiques et d'automatisation pour capturer des surfaces plus grandes. Les défis incluent le positionnement précis de la lumière, l'évitement de collisions, l'intégration fluide des systèmes et la qualité des images RTI. Nous avons étudié ces défis, développé un système d'acquisition robotique novateur et démontré ses avantages.

Reflectance Transformation Imaging (RTI) is a digital imaging technique that captures the way a surface reflects light coming from different angles. It is commonly used to study cultural heritage artifacts, such as ancient manuscripts, coins and sculptures, because it can reveal detailed surface features that may not be visible under ambient light. RTI works by capturing a series of images of an object being illuminated from different points on a hemisphere or dome. These images are then combined using specialized software to create a single, high-resolution image that encodes the surface's reflectance properties. The resulting RTI image can then be interactively visualized on a computer, allowing the user to adjust virtually the lighting direction to highlight different features of the surface.Acquisition, modeling, and visualization of reflectance of complex surfaces is still an active area of research in the RTI field. Complex surfaces can include objects with varying size, shape, material properties, which require specialized and adaptive techniques to accurately capture their reflectance properties. In this thesis, we have focused on addressing the challenges in realizing surface adaptive RTI and automating the acquisition process. We have developed methods for estimating the optimal light configuration for capturing RTI sequence adaptive to the surface being digitized. We also developed methods for stitching together multiple RTI data sets and thus improve the resolution of the RTI data. These methods are developed to improve the accuracy and efficiency of surface adaptive RTI, and to bring advances in the field of digital imaging for cultural heritage and applications. In the current state of the art, RTI acquisitions are typically carried out by manual placement of a light at different directions (free form) or use of RTI domes with fixed light positions or mechanized dome with movable light source to capture a series of images. Manually positioning the light source is a time-consuming process and lacks accuracy, repeatability. RTI domes are efficient and more reliable, however they are limited to acquisition of smaller sized objects only. To address the limitations pertaining to free form and the dome systems, we investigated the use of robotic arm and automation to streamline the RTI acquisition process. This involves the use of robotic arm to position the light source, use of a XY stage to position the surface as well as automated image capture systems. There are several benefits to automating RTI acquisition. One advantage is the ability to capture RTI images of large surfaces that are generally difficult (or impossible) to acquire using RTI domes. There are several challenges associated with the automation of RTI acquisition process using robotic arm and XY platform such as building the control systems that can accurately and reliably position the light aligning it to the required angles, collision avoidance in robotic arm planning, integration of these systems into a cohesive and user-friendly workflow, ensuring that the resulting RTI images are of high quality and meet the needs of the user. We studied these challenges in our work, built a fully functional novel robotic arm-based acquisition system and demonstrated the advantage of this system over the other existing systems.