| Explanatory text | GIS are databases that include a spatial component, i.e. that describe the location associated to a data item. A data item can be linked to geometric features such as a point, a polygon or even a 3D object such as a building. GIS are intensively used for urban and landscape management. They need various kinds of data including 2D and 3D data for natural phenomenon simulation, spatio-temporal data representation at different scales, geometric visual representations. In this project, we focus on 3D spatial data mainly in urban context and we try to address the 3G GIS construction and updating problem using ground-based videos. Mobile tools are extensively used to determine the position of their users, not only based on dedicated embedded device like GPS but also by using embedded cameras and their computation capacity. They allow thinking of real-time superimposition of 3D GIS data on video images, a technique known as outdoor augmented reality. The main challenge of augmented reality, especially on smartphones with limited CPU and memory is pose computation, i.e. the computation of both the position and the orientation of the smartphone with enough precision and low latency. The goal of this PhD is make significant improvement in the camera computation pose problem in urban environments making use of existing data (urban GIS). Applications are twofold. First, if real-time pose computation is achieved, urban augmented reality will be made possible on mobile devices, allowing people to see urban data on site (including simulation data such as energy consumption of buildings or underground pipes). Second, with or without real-time being able to register a video to an urban database will enhance the process of 3D building extraction for enriching and/or updating 3D GIS. |