Research On Rainy Traffic Sign Detection And Recognition System For Intelligent Connected Vehicle

In recent years,with the rapid development of science and technology and the national economy,people's living standards are increasing,and the number of cars is increasing year by year.The research on intelligent transportation systems and advanced driving assistance systems has attracted more and more attention.Unmanned driving has become the main development trend.The traffic sign detection and recognition system is an important part of the unmanned driving system.Through accurate recognition of the traffic signs on the road,the recognition results are fed back to the upper driving control strategy system.The vehicle responds accordingly according to the driving strategy to improve road safety,better assist unmanned driving system.Therefore,in-depth research on the traffic sign detection and recognition system has important theoretical significance and commercial value.In addition,under complex weather conditions,such as rainy weather,the traffic sign detection and recognition system will cause a decrease in the detection and recognition success rate due to weather conditions.In actual life,rain and snow weather are inevitable factors,so it is very practical application value.This article focuses on the problem of detection and recognition of traffic signs in rainy weather.In this paper,we present a rainy traffic sign detection and recognition algorithm for intelligent connected cars.We design a rainy traffic sign detection and recognition system based on the algorithm.In the system,we preprocess the rainy traffic sign based on the optimized adaptive color threshold segment and shape feature to detect the signs and use SVM to recognize the signs.The main work in the paper are as follows:(1)Because the existing public data sets have none rainy traffic,according to the research needs,this paper builds a rainy traffic sign collection platform to collects and builds a rainy traffic sign detection database.The database includes rainy traffic sign images and non-traffic sign images,and the size of each image is 864 * 480.(2)We present a set of raindrops removal methods based on image restoration.According to the morphological feature of the raindrops,we use Hough transform to detect raindrops in the image.According to the principle of image restoration,decompose the image into structure image and texture image.Using the raindrops domain information to restore the structure image and texture image,which solves the impact of raindrops on the windshield on the detection of traffic signs during image acquisition,and obtains a better image of raindrop removal.(3)This paper proposes a rainy traffic sign detection algorithm based on optimized adaptive threshold.The traditional threshold segmentation method based on different color spaces can effectively segment the traffic sign area of interest.However,under rainy conditions,the background brightness of the image is different,and the contrast of the sign decreases,making it difficult to determine the range of color threshold.To solve this problem,we present an optimized adaptive color threshold method: Firstly,the contrast of the region of interest of the image is improved by means of color enhancement;Then grayscale the image,calculate the image gray probability,and the cumulative distribution under different gray values is calculated and construct a function;Finally,an adaptive threshold determined by the probability density and cumulative distribution function is used to segment the image with a color threshold to obtain an image containing the region of interest.The geometric features of traffic signs are used to accurately detect images.The detection rate of the algorithm in the rainy day traffic sign data set is 81.46%,and a series of experiments verify the feasibility of the algorithm.(4)This paper designs a traffic sign recognition algorithm based on SVM.We introduce the extraction process of HOG feature and Gabor feature.According to the principle of SVM,we design and train a single feature SVM classifier and a fusion features SVM classifier.We obtain the optimal feature vector dimension in the experiments.And using the optimal parameters to train the SVM classifier.After the experiments in the public data set,the recognition rates are 92.19% and 98.16%,we get conclusion that the fusion features SVM classifier has a higher classification recognition rate.The conclusion proves the effectiveness of the fusion feature.In summary,this paper studies the rainy day traffic sign detection and recognition algorithm,builds a traffic sign collection platform,propose a complete algorithm from image preprocessing to sign detection and recognition,have been verified in the data set to achieve the detection and recognition of the rainy traffic signs.