Aiming at problem of low recall rate and high error rate of traditional traffic sign detection algorithm based on integral channel feature（ ICF） and Adaboost

propose a two-stage traffic sign detection method. In the first stage

using spectral clustered integral channel features obtained by spectral clustering are combined with Adaboost to learn the overall detection model

which is applied to the input image to obtain region of interest（ ROI）. In the second stage

histogram equalization is imposed on ROI

and then a shape classifier using support vector machine（ SVM） is employed to filter candidate object regions obtained in the former stage to remove the false positives. Experimental results show that the proposed SICF-Adaboost + SIFT-SVM detection algorithm method built upon a cascade classifier framework possesses higher detection rate compared with the traditional ICF Adaboost detect algorithm in dealing with high light intensity

motion blur

fog

and noisy similar object

which is suitable for traffic sign detection in real-world scenes.