Skip to Main content Skip to Navigation

Analyse morphologique d'images pour la modélisation d'environnements urbains

Abstract : Urban modeling allows the development of many applications such as: 3D navigation, urban planning, scene modeling for the entertainment industry, etc. The technological challenge is to increase the productivity and the realism of urban modeling. This thesis is developed in the framework of Cap Digital Business Cluster TerraNumerica project. The project aims at developing a production and exploitation platform, by allowing the definition and visualization of synthetic urban scenes. In this context, our main contribution to the project consists in segmenting color images and point clouds in order to assist urban modeling. This document is mainly divided into three parts. In the first part, we have shown the ultimate opening benefits as a generic operator of segmentation. Then we have proposed several improvements to make it more robust to masking and leakage problems. The performance of these improvements is illustrated in our study framework, scene-text detection and cell segmentation. The second part is focused on façade analysis. Façade modeling is performed on the scope of a single building. However, during the image acquisition process, several buildings appear in the same image. We propose an automatic method to separate different façades included in the image. Then, we focus on the semantic extraction from the façade. It consists in segmenting it by floors, windows, balconies... to provide a realistic building model. The third part of this thesis is focused on point cloud analysis. In urban modeling, it is necessary to introduce some elements such as street furniture and pavement. We present tools for the detection and classification of artifacts. These tools allow: 1- the filtering of data in order to facilitate the modeling process and 2- the re-introduction of some elements (lampposts, sign boards, bus stop, etc), improving visual realism in the modeled scenes. We also propose an automatic method for pavement segmentation.
Document type :
Complete list of metadata
Contributor : Ecole Mines ParisTech Connect in order to contact the contributor
Submitted on : Tuesday, May 4, 2010 - 8:00:00 AM
Last modification on : Wednesday, November 17, 2021 - 12:27:08 PM


  • HAL Id : pastel-00005974, version 1


Jorge Eduardo Hernández Londoño. Analyse morphologique d'images pour la modélisation d'environnements urbains. Mathématiques [math]. École Nationale Supérieure des Mines de Paris, 2009. Français. ⟨pastel-00005974⟩



Record views


Files downloads