Fast greyscale road sign model matching and recognition

Mobile Mapping is a standard technique for compiling cartographic information from a mobile vehicle. This paper proposes a novel method for modelling the recognition in a Mobile Mapping process that consists in fitting a model to recover the sign distortion and applying recognition techniques on weak classifiers cascade results. The images received from Adaboost learning algorithm with weak classifiers cascade are processed to capture the sign and to perform the following recognition. Radial symmetry, false contours extraction techniques, and ellipse fitter algorithms are used to solve the problem of signs distortion. The procedure is robust even in the case of high variance of sign appearance as noise, affine deformation, and reduced illumination.