ADAS (Autonomous Driving Assistance Systems) are more and more integrated in vehicles and provide to drivers more confort and safety. In this thesis, we focused on Intelligent Speed Adaptation. We proposed an approach combining vision and navigation in order to optimally manage the driving context information.Roadsigns, subsigns and markings are visual data used by the driver to determine the current temporary speed limitations. Many research were conducted during last years to recognise the first one, contrary to the second. Commercialised products are even implemented in vehicles. We thus developped a subsign detection and classification module using greyscale images. A morphological reconstruction with a growing region helped us to focus the segmentation on highly contrasted pixels surrounded by homogeneous regions. Global descriptors such as PHOGs combined to a hierarchical structure of SVMs were then used to classify the output rectangles. Finally, we eliminated subsigns which are not applicable to the current lane by considering markings.After having developed a vision module integrating all the available information, we improved the navigation system. The objective was to extract from an embedded database the driving context related to the vehicle position. Urban context or not, functional class, road type and speed limit were collected and modelised into criteria. The sensor reliability was then computed and depended on the satellite configuration and the network digitisation quality. Confidence in each speed limit combined all these elements.Fusion of both sources with the Dempster-Shafer theory led to very good performances on our databases et showed the importance of all the used information.
