Robust Implementation for Circular Traffic Sign Recognition

The design of traffic sign recognition (TSR) system has been a challenging practical problem for many years. The frequently used approaches involve color segmentation, shape analysis, and pictogram classification. To cope with the requirements of robustness and accuracy, we introduce a robust circular traffic sign detection approach consisting of a color enhancement method which utilizes the nature of Lab color space and a circle detector based on improved constrained fast radial symmetry transform. To effectively represent the pictogram of candidate sign, an informative data representation method, i.e., circular local binary pattern histogram, is presented for designing the traffic sign classifier through support vector machine. The system performance under extensive experiments has shown robust detection effects and outperforms existing approaches in terms of classification accuracy.