Context
In a world facing profound upheavals—ecological, energy-related, economic, health-related, social, and more—territories are at the heart of the most complex decisions. The stakeholders within these territories are under increasing pressure to anticipate, adapt, and invent new approaches to planning and development. It is now essential to be able to anticipate territorial evolution and simulate different management scenarios in order to assess, and even compare, their impacts. This is the objective of the Digital Twin of France and its Territories (JUNN) project, initiated and co-led by the IGN (National Institute of Geographic and Forest Information), Cerema (Center for Studies and Expertise on Risks, Environment, Mobility and Urban Planning), and Inria (National Institute for Research in Digital Science and Technology).
The JUNN project will leverage an unprecedented amount of remote sensing data on the entire French territory to process and analyze. In particular, the airborne LidarHD campaigns and satellite-based Digital Surface Models will provide spatio-temporal 3D data that will be used, not only for updating 3D city models over time, but also for better understanding the evolution of urban landscapes during long term periods. In this context, designing methods for efficiently analyzing the geometric changes and understanding the evolution of urban attributes such as urban growth and architectural variations constitutes a key scientific challenge.
Objectives
The goal of this PhD is to (i) develop efficient methods for detecting 3D changes and expressing them with simple geometric shapes, and (ii) analyze the spatio-temporal distribution of these geometric changes at large scales (i.e. from city districts to the entire country).
In contrast to existing 3D change detection methods that mostly operate from 3D point clouds, e.g. [1,2], the PhD candidate will investigate, as first objective, change detection methods that directly operate from more concise geometric primitives such as planes and 3D polygons. This strategic choice is motivated by both efficiency reasons as point-based methods suffer from a low scalability and interoperability reasons as such geometric primitives will directly feed the building reconstruction methods of the JUNN project for efficient 3D model updates. The candidate will investigate methods for detecting planar variations in a pair of point clouds. One possible solution will be to adapt static mechanisms such as [3] by using similarity metrics between planar shapes, as proposed in [4] for 3D data registration. The candidate will also investigate data structures to efficiently organize and parse the detected planar changes, e.g. by using Level of Detail trees [5].
The second objective will be to evaluate the potential of these detected spatio-temporal variations for understanding evolution of urban attributes. In particular, the PhD candidate will develop models for analyzing the spatio-temporal distribution of planar shapes at large scales and seek potential correlations on a variety of attributes that characterizes the city evolution in terms of shape, physics or functionality. A first naïve approach will be to extend the statistical models developed in [6] for basic 2D building footprints with more expressive 3D planar primitives.
Keywords
Geometry processing, 3D computer vision, machine learning, statistical analysis, urban reconstruction, planar shape detection
References
[1] Stilla and Xu. Change detection of urban objects using 3D point clouds: A review. P&RS journal, 2023
[2] de Gélis, Lefèvre and Corpetti. 3D urban changes detection with point cloud siamese networks. ISPRS archives 2021
[3] Yu and Lafarge. Finding Good Configurations of Planar Primitives in Unorganized Point Clouds. CVPR 2022
[4] Li and Lafarge. Planar Shape Based Registration for Multi-modal Geometry. BMVC 2021
[5] Pan, Zhang, Liu, Gong and Huang. Building LOD Representation for 3D Urban Scenes. P&RS journal, 2025
[6] Zhu et al. GlobalBuildingAtlas: An Open Global and Complete Dataset of Building Polygons, Heights and LoD1 3D Models. ArXiv 2025.
More info on the position can be found at https://team.inria.fr/titane/files/2026/03/sujet_JNFT_spatiotemporal_analysis.pdf and on the JUNN project at https://team.inria.fr/titane/the-jnft-project-2026-2030/
