Research On Traffic Sign Detection And Recognition Algorithm

Traffic signs are one of the most important road messages that use different combinations of shapes,patterns and colors to convey different road information to drivers and pedestrians.The intelligent vehicle system can effectively reduce the incidence of traffic accidents by detecting the current traffic sign information,reminding the driver of the warning or warning.Therefore,the research on traffic detection and recognition system has important practical significance.In this paper,we will implement a traffic sign detection and recognition algorithm with good detection rate and recognition rate.In the traffic sign detection stage,this article mainly use of traffic sign color and shape characteristics.In traffic sign color extraction,this paper respectively uses RGB color enhancement method and color classifier method to carry out traffic sign color extraction.RGB color enhancement method using the normalized color space,calculate the difference between channels to achieve color enhancement.The method of color classifier trains the SVM color classifier by extracting the red and blue pixels under different conditions,and sends each point of the image to be detected to the classifier to obtain the gray feature image.In terms of traffic sign candidate area extraction,this paper uses the traffic sign shape information,using template matching and maximally stable extremal regions.Template matching is the use of the template design and gray image convolution calculation,the convolution value of a large area is the potential area of traffic signs.In the region of the maximum stable extreme value,the gray level of the extracted gray level map is relatively stable,while the change of the gray level value of the traffic sign region after color extraction is relatively stable.The maximum stable extreme value region detected can be regarded as the traffic Sign candidate area.In this paper,we use the HOG feature combined with SVM classifier and convolutional neural network respectively to screen the candidate regions of traffic sign to determine whether it is a traffic sign,so as to realize the location detection of traffic sign.In the stage of traffic sign recognition,this paper designs a traffic sign recognition method based on HOG features and SVM classifier from coarse to fine using the traditional machine learning method.The method first determines the major categories of traffic signs,and then uses the category classifier to obtain a specific category.In addition,this paper designs a 7-layer convolutional neural network to recognize traffic signs and reduces the parameters of the convolutional layer by dividing of convolution kernels.In this paper,the proposed algorithm is tested.Through experiments,the algorithm in this paper is effective in traffic sign detection and recognition.