Research On Detection And Recognition Of Road Traffic Signs

In recent years,our country's economy has developed rapidly,cars are becoming more and more popular in people's daily lives,and people are increasingly demanding safe and fast intelligent transportation systems.The traffic sign is a key facility of the transportation system and plays an important role in regulating people's traffic behavior and indicating road conditions.Therefore,the detection and recognition of traffic signs is a key point for realizing the intelligent traffic system.Accurate positioning and identification of traffic signs in roads is essential for real-time analysis of road conditions and for providing drivers and intelligent traffic systems with the necessary traffic information.However,traffic signs in actual road scenes are easily disturbed by external and internal factors such as light,weather,obstruction,deformation,fading,etc.,which causes great troubles for detection and recognition.In addition,due to the real-time requirements of traffic information,the algorithm proposed by us seeks higher accuracy while also taking into account the execution speed of the algorithm,which puts forward higher requirements for our detection and recognition process.This article has conducted in-depth research on the characteristics of road traffic signs and the possible interference in natural scenes.Combined with the research results of domestic and foreign scholars,a traffic sign detection method based on maximum stable extreme value region(MSER)algorithm and convolutional neural network(CNN)was proposed.In addition,this paper proposes an improved CNN network model to complete the recognition of traffic signs.The main work of this paper is as follows:(1)In-depth analysis of the influence of color,shape characteristics and environmental factors of traffic signs on road images,a traffic sign detection method based on MSER+CNN is proposed.Taking into account the unique color characteristics of traffic signs,a specific color in the road image is enhanced to obtain a grayscale image after the traffic sign area is enhanced,and then a stable region in the enhanced image is detected using the MSER algorithm to obtain a region of interest(ROI).Then based on the traffic signs width,height,area,aspect ratio and other shape characteristics of the ROI for a preliminary screening.Finally,a CNN model is trained to perform secondary discrimination on the preliminary filtered ROI,and the exact location of traffic signs in the road image is finally obtained.(2)For the recognition of traffic signs,this paper first uses the histogram equalization and image normalization to preprocess traffic signs.Then analyzing the advantages and disadvantages of the currently used CNN model for image classification.Finally,the LeNet-5 model is selected to complete the traffic sign recognition.By analyzing the experimental results of the LeNet-5 model,the problems of the classic LeNet-5 network in identifying traffic signs are pointed out.Then the improved CNN network model is proposed,and the improved network model and other recognition methods are compared and analyzed for the recognition rate and recognition time of traffic sign recognition.Experimental results show that the improved network model in this paper has some advantages over other classical network models and feature extraction + classifier methods in terms of recognition rate and real-time performance.