| Urbanisation still remains one of the main problems worldwide. The extent and rapidity of the urban growth induce a number of socio-economic and environmental conflicts everywhere. In order to reduce these problems, urban planners need to integrate spatial information in planning tools.;The aim of our study is to develop a new method for semi-automatic building extraction with VHSR. The different steps performed to achieve our objective are each presented in a chapter.;In the first chapter, the general context of our research is described with the definition of our objective. After a short historical review of urbanisation, we focus on urban growth and associated problems. In the following we discuss the possible contributions of geography to reduce these problems. After discussing concepts, theories and methodologies of geographical analysis in urban areas, we present existing general urban planning tools. Finally, we show the special interest of our study that is due to a growing need to integrate spatial information in these decision support tools.;In the second chapter we verify the possibility of reaching our objective by analysing the technical characteristics of the images, the noise and the distortions which affect the images. Quality and interpretability of the studied image is analysed in order to show the capacity of these image to represent urban objects as close to reality as possible. The results confirm the potential of VHSR Imagery for urban objects analysis.;Actually high expectations are made on Very High Spatial Resolution imagery (VHSR). These high-spatial resolution images are available at a reasonable price and due to short revisit periods, they offer a high degree of actuality. However, interpretation methods seem not to be adapted to this new type of images.;The third chapter deal with the preliminary steps necessary for the elaboration of our method of building extraction. First, we evaluate the quality of the Sherbrooke Ikonos image (geometric and radiometric quality), the basic image of our analysis. A review of existing methods clearly show a common limit: the detection of building boundaries. Consequently, we evaluate the efficiency of several segmentation methods that finally induces a change in our methodological approach.;The fourth chapter contains the central part of our work, which consists in the development of a building extraction method. After strict formalisation of our, objectives, we present the theoretical principles of our approach based on textural buildings definition. In the identification process we use only one parameter that accounts at the same time for the variance of the building and the variance of its immediate surroundings. In the following, additional information (shadow and vegetation) is integrated to reduce commission errors.;The last chapter exposes the results. They clearly show the capacity of our method for building identification. However, they show some limitations of application, especially on large size buildings and/or in heterogeneous areas. We also propose possible applications such as analysis of suburban buildings or detection of natural disaster damages.;The main outcome of this work is the originality of our theoretical approach that encourages new reflections for future research. |
