Research On Image Enhancement And Recognition Technology Of Highway Signs In Foggy Weather

Street sign recognition system is the key techniques in modern intelligent driving,which assists drivers to obtain sign information accurately.It avoides the distraction of road sign information when drivers are unfamiliar with road conditions,and also contributes to driving safety.It will increase the risk of highway traffic tremendously unless identifies the accurate information in time,especially driving on the highway at high speed.In addition,the image quality collected by imaging device in foggy weather is relatively poor,which not only makes difficulties for recognition system to process accurately,but also affect the recognition precision.In view of these problems,we study the image enhancement and recognition technology of foggy highway in this paper.This paper focuses on image enhancement,street sign character segmentation,and street sign character recognition.In terms of image enhancement,we study the dark channel prior enhancement algorithm,Retinex-based enhancement algorithm and histogram equalization enhancement algorithm,and compare the experimental results of these enhancement algorithms.Through the subjective and objective evaluation,the most effective selection dark channel prior algorithm is used as the image enhancement method in this paper;In terms of the character segmentation,we use the method based on the combination of MSER and projection to process the enhanced image,the segmentation algorithm of this paper is compared with the MSER method and the projection method.The data shows that the segmentation algorithm in this paper has the best effect,and the segmentation rate reach 95.9%,this method can not only solve the problem of Chinese characters structure nesting in MSER method,but also solve the problem of illumination influence under natural conditions;In terms of character recognition,we study template matching algorithm,support vector machine algorithm and convolutional neural networks algorithm,then compare these algorithms with experiments,The results shows that the convolutional neural networks has the most effective recognition for a large number of highway sign characters,which attains the rate to 92.8%,so we choose convolutional neural network for character recognition in this paper.In order to verify the feasibility of the model,this paper conducts experiments on highway signs collected in foggy weather,and evaluates the effect of image enhancement and performance of street sign characters recognition through experimental data.This paper compares the recognition rate of two segmentation algorithms combined with convolutional neural networks and the recognition rate before and after image enhancement.The data shows that the segmentation algorithm combined with the convolutional neural network can reach a recognition rate of 92.8% in foggy weather,which fully proves that the segmentation algorithm can effectively improve the holistic recognition rate and the necessity of image enhancement in foggy weather.The model in this paper provide safe and stable guarantee for intelligent driving system.