[HDR] Segmentation and 3D modeling of urban scenes from point clouds

by Tania Landes

3D modeling responds to both an economic and an environmental issue, whether at the building or at city scale. Over the past ten years, acquisition techniques have evolved considerably in terms of their speed, the amount of data to be managed, the heterogeneity of information acquired by multi-sensor systems, as well as in terms of processing methods. In this context, new processes such as the “scanto-BIM” process have emerged. It characterizes the steps leading from point cloud to an intelligent digital model called BIM (Building Information Modeling). By adopting the digital model, building stakeholders are able to perform simulations and reduce, in addition to costs, the environmental impact related to interventions on the building, throughout its life cycle. In practice, to achieve an intelligent digital building model from a survey, many technological obstacles need to be overcome. In this context, I had the chance to supervise various research works on the topics of 3D data acquisition, from the processing of this data to the production of the 3D digital model.

Through doctoral research studies that I had the opportunity to co-supervise, I participated in the construction of processing chains on point clouds acquired by airborne LiDAR, terrestrial LiDAR and other 3D sensors. This work has led to the development of innovative acquisition systems (PhD Mittet, 2015), registration algorithms (PhD Hullo, 2013 and PhD Lachat, 2019), segmentation and modeling of point clouds for the reconstruction of buildings (PhD Tarsha-Kurdi, 2008), of facades (PhD Boulaassal, 2010) and of indoor scenes (PhD Macher, 2017). The issue of assessing the quality of models was addressed in every thesis, particularly at the scale of buildings (Mohamed thesis, 2013).

Based on the experience acquired through all of this work, we focused also on environmental issues, with the ambition to reconstruct urban trees in 3D. This daring challenge was met, thanks to the collaboration of specialists in urban climatology from the ICube-TRIO team. Finally, a new topic of research has been created within our team: the contribution of laserscanning to urban climate modeling. Urban climate simulation models need truthful 3D models of trees in order to assess the impact of vegetation on the creation of urban microclimates. A first work has shown that the challenge can be faced at the tree level and a park level (PhD Bournez, 2018). The study is currently continuing at city and regional levels (PhD Philipps).

In the first part of this work, we will recall the major advances in each step of the established processing chain, before proposing research perspectives. In the « scan-to-BIM » process, the segmentation, up to now mainly geometric, deserves to be enriched with semantic information in order to result in a socalled « digital twin » of the building. The efforts that we will carry out in this direction will also benefit urban climatological models, which require knowledge of urban geometry. To control the errors affecting the models delivered by the processing chain remains a center of interest.

In view of our skills and the growing interest of our European countries for a green economy and for energy transition, the future offers good prospects for the development of our team’s research topics.


The manuscript of the HDR can be accessed here.