Abstract—This paper presents an automation approach towards the detection and classification of cracks on bridge surfaces using a robot platform. The approach is designed to exploit the physical features of cracks and is therefore capable of overcoming the challenges that traditional crack detection approaches are faced with. The approach adopts the Beamlet and Wavelet Transforms in the realization of a robust crack segmentation scheme. The Radon transform is coupled with the Projection Variance towards the extraction of crack features which facilitates a high specificity even in the presence of noise and texture irregularities. Finally, in order to render all this information useful and applicable towards the maintenance of bridges, a classification scheme is proposed which classifies cracks into non-crack, simple crack and complex crack categories. The classification scheme is realized through an AdaBoosted RVM implementation that achieves a high classification accuracy and generalization. This detection and classification system is deployed on the six-legged robot platform designed to operate semi-autonomously on bridges. The performance of this scheme is verified through comparison experiments with state-of-the-art and the experimental results indicate that the proposed scheme achieves effective results while outperforming some of the state-of-the art in terms of accuracy and classifier training time.