Trajectography registration by combined datamodel deformation for geometric enriching of existing city models ; Recalage conjoint de données de cartographie mobile et de modèles 3D de bâtiments

For many years, mobile land scanning vehicles have been developed to simultaneously acquire highly accurate laser data and high-resolution georeferenced images. A major application of these data is to exploit their very high level of detail (LOD) to enrich the 3D geographic databases built from aerial images and therefore much lower LOD. The 3Dgeographic databases and mobile terrestrial data prove to be very complementary: roofs are seen from the air but not on land, and facades are very poorly seen from the air but very precisely on land. Geographic databases consist of a set of geometric primitives (3D triangles) of coarse LOD, but present the advantage of being available over large geographical areas. Mobile mapping vehicles offer much more partial coverage but guarantee very fine LOD data. These vehicles also have limitations: in urban environments, the GPS signal needed for good data georeferencing is liable to being disrupted by multi-paths or even stopped during GPS masking phenomena linked to narrow streets or high buildings. The GPS sensor no longer picks up enough satellites to accurately determine its spatial position. These complementary data each have their own geo-referencing and geolocation uncertainties of drift, ranging from a few centimetres to several metres. This means that different datasets in the same area do not coincide. That is why a realignment is essential to bring this highly detailed mobile data into line with the less detailed geographical databases.In this thesis, we have finely modelled all the sources of uncertainty involved in boththe process of building the lidar point cloud and the geographic database to jointly (simultaneously) re-align the data between them. This work around uncertainties makes it possible to model them, then to exploit them in the realignment process, and finally to to propagate them on the final product, by means of a Gauss-Helmert model. The process is based on an Point to plane ICP (Iterative Closest Point) method. This realignment simultaneously ...