Phd defense of Flavien ARMANGEON

Flavien ARMANGEON

Scene-Functional Alignment In An Industrial Environment

Abstract

The objective of my thesis is to link 2D and 3D acquisitions of an industrial scene to the functional schematics using tools drawn from data science and computer vision. This alignment is useful for many applications such as building a 3D model linked to the functional information, which can be used to train operators in a virtual environment. The main difficulty is that the schematics have no distance information between objects. Thus, the scene-functional alignment must rely purely on the functional relationships between pieces of equipment. The problem has been divided into three steps: segmenting the scene to identify the objects to be matched, defining a relevant and shared representation for both the acquisitions and the functional data, and finally aligning the scene with the functional schematics using this representation. I first focused on visibility estimation in a point cloud, a key step in our segmentation approach. Subsequently, I studied different representations and selected a graph-based approach in which the nodes represent pieces of equipment and the edges represent functional links between them. The idea is to extract a graph from the scene and another from the schematics using the same conventions. Finally, I explored several alignment methods to match the nodes from the two graphs and proposed a variant adapted to our problem. This is the first end-to-end approach for aligning an industrial 3D scene with its functional information.

PhD supervisors

• Rafael GROMPONE VON GIOI, Chercheur, Centre Borelli, ENS Paris-Saclay
• Gabriele FACCIOLO,  Professeur, Centre Borelli, ENS Paris-Saclay
• Enric MEINHARDT-LLOPIS,  Chercheur, Centre Borelli, ENS Paris-Saclay
• Thibaud EHRET, Chercheur, Centre Borelli, ENS Paris-Saclay
• Guillaume THIBAULT, Ingénieur R&D EDF
• Marc PETIT, Ingénieur R&D EDF

Defense committee

• Raphaëlle CHAINE, Professeure, LIRIS, Université Claude Bernard Lyon 1 (Rapporteure & Examinatrice)
• Jean-Emmanuel DESCHAUD, Professeur, CAOR, Mines Paris – PSL (Rapporteur & Examinateur)
• Julie DIGNE, Directrice de recherche, LIRIS, Université Claude Bernard Lyon 1 (Examinatrice)
• François GOULETTE, Professeur, U2IS, CAOR, ENSTA Paris (Examinateur)
• Loic LANDRIEU, Directeur de recherche, IMAGINE, IGN in the LASTIG lab, ENPC (Examinateur)