Road sign recognition using support vector machines and histogram of oriented gradients

In this paper, we consider the recognition of traffic signs using support vector machines (SVMs) and features based on histograms of oriented gradients (HOG). We approach the training of classifier with two well developed multiclass support vector machine formulations proposed earlier by Weston & Watkins and Crammer & Singer. Feature space straightening is approached with Jensen-Shannon and histogram intersection kernels. Due to computational efficiency reasons we propose the use of the homogeneous kernel mapping presented recently. The comparative study based on the German Road Traffic Sign Recognition Benchmark dataset shows the effectiveness of our approach.