This work presents a semi-automatic approach to the 3D reconstruction of Heritage-Building Information Models from point clouds based on machine learning techniques. The use of digital information systems leveraging on three-dimensional (3D) representations in architectural heritage documentation and analysis is ever increasing. For the creation of such repositories, reality-based surveying techniques, such as photogrammetry and laser scanning, allow the fast collection of reliable digital replicas of the study objects in the form of point clouds. Besides, their output is raw and unstructured, and the transition to intelligible and semantic 3D representations is still a scarcely automated and time-consuming process requiring considerable human intervention. More refined methods for 3D data interpretation of heritage point clouds are therefore sought after. In tackling these issues, the proposed approach relies on (i) the application of machine learning techniques to semantically label 3D heritage data by identification of relevant geometric, radiometric and intensity features, and (ii) the use of the annotated data to streamline the construction of Heritage-Building Information Modeling (H-BIM) systems, where purely geometric information derived from surveying is associated with semantic descriptors on heritage documentation and management. The “Grand-Ducal Cloister” dataset, related to the emblematic case study of the Pisa Charterhouse, is discussed.

Dépositaire des ressources numériques des grottes et abris ornés, le Centre national de la préhistoire (CNP) s’attache à encourager et à réaliser en moyenne une campagne de numérisation 3D par an, afin d’enrichir ce fonds de données. Ces ressources constituent à la fois des archives des sites d’art pariétal et des supports pour leur conservation, leur étude et leur médiation. Du fait qu’elles n’ont subi aucun traitement ou manipulation susceptibles de biaiser leur exploitation, les données brutes constituent pour le CNP et ses partenaires une ressource précieuse pour l’étude et la documentation des sites ornés. Ainsi, il devient aujourd’hui impératif de s’intéresser à la documentation et la formalisation de ces ressources, afin de les pérenniser et garantir leur utilisation au bénéfice du plus grand nombre, qu’il s’agisse d’experts ou du grand public. À ce titre, le CNP a sollicité l’UMR MAP (Modèles et simulations pour l’architecture et le patrimoine) pour initier ensemble dans le cadre de l’accord-cadre CNRS-ministère de la Culture un travail d’une part sur la description de ces données et d’autre part sur leur diffusion. Le premier point implique de produire des descripteurs pertinents permettant d’accompagner les données 3D d’indications claires quant à leur provenance, leur nature et leur qualité. Le second aspect implique l’harmonisation et la pérennisation des données, ainsi que le développement d’un prototype de plateforme web permettant leur visualisation et manipulation à travers une interface adaptative. Ce travail s’inscrit dans une réflexion plus globale dont l’objectif est de contribuer à la traçabilité des données numériques 3D destinées à des utilisations scientifiques. Nous proposons dans cet article de détailler l’ensemble du travail réalisé dans le cadre de ce projet sur les ressources numériques des grottes ornées, en s’intéressant particulièrement aux développements réalisés pour la visualisation des données 3D.

The French collection of Plans-Reliefs, scale models of fortified towns, constitutes an awesome architectural heritage. Facing this patrimonial richness, many cities, represented on these plans-reliefs, would like to expose, develop and exploit this historical heritage. However, the fragility, the dimension of the supports and the exposure conditions makes this acquisition difficult. Thus, the creation and the exploitation of a virtual model constitute an interesting alternative. This paper presents a method exploiting historical documentary for the 3D semantic modelling and the valorisation of the virtual Plan-Relief of Verdun. The original Plan-Relief is currently enclosed in containers in Paris.

Despite the progress in three-dimensional scanning, some architectural artifacts remain a digitizing challenge. Scale models and more especially the plans-reliefs of Louis XIV of France have specific characteristics such as size, scale, number, etc. A knowledge-based modeling approach is developed to address the limitations of digitizing tools. Our study deals with the fortified areas of the scale models. Bastioned fortification works extend over wide areas but they are built according to design rules. Once studied and organized, fortification knowledge has been used to create a library of parametric components. Implemented in Grasshopper, the components were manually adjusted to different practical cases. The library was then validated and we are now focusing on the automation of the adjustments of the components. Thereupon a reverse-engineering approach has been set up. Semantic segmentation algorithms have been defined and implemented in Grasshopper to automatically extract fortification features from 3D surveys based on the knowledge of fortification design rules. Along with the three-dimensional modeling of fortification, an automatic reconstruction of the city parts of the scale models is underway. Both these studies are part of a project aimed at valorizing and diffusing a very unique cultural heritage collection. As such, knowledge models are precious assets both the digitizing and the semantic enhancements of the final application.

Keywords: 3D Modelling, Scale Models, Photogrammetry, Laser Scan, Segmentation Urban scale models depicting whole towns such as the hundred-scale model collection known as plans-reliefs are a valuable source of information of cities and their surroundings. These physical representations of French strongholds from the 17th through the 19th century suffer from many problems that are, among other things, wear and tear or the lack of visibility and accessibility. A virtual collection would allow remote accessibility for visitors as well as history researchers. Moreover, it may also be linked to other digital collections and therefore, promote the collection to make people come to the museums to see the physical scale models. We also work on other physical town scale models like Epinal for which the scale is a bit higher. In a first part, we define a protocol for acquiring 3D meshes of town scale models from both photogrammetric and scanning methods. Then we compare the results of both methods The photogrammetric protocol has been elaborated by choosing the most accurate software, 123DCatch, which asks for about 60 pictures, and defining the settings needed to obtain exploitable photographs. In the same way, we defined the devices and settings needed for the laser scan acquisition method. In a second part, we segment the 3D meshes in planes by using Geomagic, which has been chosen between several programs, for its accurate resulting geometry.