Road signs play a significant role in traffic control and road safety. Road signs have to be enough salient so that they attract the driver's attention. In this thesis, we study the feasibility of an algorithm to automatically estimate the saliency of road signs in road scene images in order to diagnose a road network by the processing of images acquired by a camera onboard a dedicated vehicle. Our seminal thinking is to rely on confidence values of a learning algorithm called " Support Vector Machines " to model the search saliency for an object of interest: a (set of) road sign(s). A statistical analysis on visual behavior data, collected by an experimental protocol in cognitive eye tracking, shows the correlation of the proposed model of search saliency of road signs with the human visual performance in near-driving situations and proof is quality in order to valuate the road sign saliency.
