A CNN-based Traffic Sign Detection and Classification Method Using Priori Knowledge

Traffic sign detection and classification is one of the main tasks of the advanced driving assistance system (ADAS). It is an integral part of the automatic driving vehicle. How to improve the accuracy and detection speed of traffic sign recognition has always been the focus of research. To solve the above problems, a fast three-stage traffic sign detection and classification method is proposed in this paper to improve the algorithm accuracy. In the first stage, we develop a probability distribution model based on the color, location, and type of traffic signs as a priori information, which can drastically minimize the search range of signs and enhance detection efficiency. In the second stage, this paper proposes an image color segmentation method based on Gaussian mixture model (GMM) as the detection module, uses the YCbCr color model for image segmentation. The morphological closure is then performed to refine the segmented image. In the third stage, the classification module classifies the extracted target areas through deep convolutional neural network (CNN).