Detection And Identification Of Traffic Signs Under Hazy Conditions

The prosperity of the automotive industry and the rise of artificial intelligence technology have enhanced the development of the Intelligent Traffic System(ITS)greatly.Intelligent Traffic System includes orientation,entertainment,emergency communications,anti-theft,and information provision,which can bring great convenience and guarantee the security when people is traveling.The automatic identification of traffic signs is one of the most important components of ITS.And it is also an indispensable pivotal key technology in driverless system and driver assistance system.The road environment is complex and changeable.Traffic signs are easily disturbed by multiple factors such as sunlight,weather,corrosion,physical obstruction,and deferent shooting angles in natural scenes.This poses high technical requirements for the detection and identification of traffic signs.Recent years,the appearance of haze weather in most cities in China has made the images collected by the system equipment fuzzy,unclear and lacking a large amount of detailed information,which leads to the fact that the images can't be detected and identified directly.This thesis studies the detection and identification of traffic signs under hazy conditions,solves the problem of haze disturbance of images,on which this thesis also realizes the rapid and accurate identification of traffic signs based.The main content of the article is divided into three parts: image fog removal under hazy conditions,traffic sign detection,and traffic sign recognition.In terms of image defogging,this thesis improves the traditional dehazing algorithm based on the dark channel prior principle.Using an adaptive global light intensity estimation method,the image is segmented into several sub-images,and the global statistical intensities of the road interference sources are eliminated by mathematical statistical characteristics.For the estimation of the transmittance,the improved algorithm introduces the color information of the pixel itself and the neighborhood as the measuring factor,and reduces the halo blur produced by the traditional algorithm at the edge of the image and the distance mutation.The effectiveness of the improved algorithm is verified by experimental comparison.In terms of traffic sign detection,this thesis first uses the principle of HSV color threshold segmentation to quickly detect the image that is more quickly,locate the candidate area of the traffic sign and eliminate most of the irrelevant information,and perform denoising on the extracted image.A corresponding SVM classifier based on HOG feature extraction is designed for the shape of the traffic sign,and the final detection and judgment of the image is performed in a combined manner.By experimental verification,the detection algorithm of this thesis makes most use of the relationship between the two characteristics of the color and shape of the traffic sign,and can meet the accuracy requirements of road traffic sign detection.In terms of traffic sign recognition,the principle knowledge and parameter methods of convolutional neural network are introduced.The classic Le Net-5 convolutional neural network model is optimized and used for traffic sign recognition.Through the contrast enhancement of preprocessing and pre-classification of the input image to improve the accuracy of the algorithm,and to adjust the network structure and parameter configuration of Le Net-5 to meet the real-time requirements,the network has faster convergence and recognition speed.Experiments have shown that the improved recognition algorithm is more accurate and achieves a recognition rate of 99.0% on the GTSRB data set,and it greatly improves the speed and can meet the real-time system requirements.