“…At present, there are many methods for traffic sign identification. Due to the unique shape and color of traffic signs, some scholars have proposed color space [2][3] , a method [4] of identifying traffic signs according to shape characteristics, or [5] of combining color and shape characteristics, or extracting candidate regions under color space to support vector machine (SVM) classifier identification [6][7] .In recent years, due to the rise of the convolutional neural network [8] , some scholars have proposed the method of detecting traffic signs based on the convolutional neural network [9] , which extracts the candidate regions through the pre-selected box and is sent to the classification network for identification.But these methods all have some drawbacks,According to the existing publicly available traffic sign dataset, we found that the world's mainstream traffic signs, whether German traffic sign data set (GTSDB),Belgian traffic sign data set (Belgium TS), or Chinese traffic sign dataset (CTSDB), are single and fixed, warning signs are circular shape of red outer circle, and blue background circular shape. Therefore, according to the color and shape characteristics of circular traffic signs, this paper proposes an algorithm that separates red and blue color channels under HSV color space, uses Hough transformation to extract the original image candidates under each channel, and sends it to the constructed shallow convolutional neural network (CNN) classifier to detect circular traffic signs.…”