| Visual perception has become one of the hot areas in the field of intelligent driving.However,under the influence of fog,rain,low light and other environmental factors,visual sensors are insufficient to obtain road traffic information,which affects the operation of intelligent driving vehicles relying on visual SLAM.Therefore,it is one of the key problems in visual SLAM to study the fast enhancement algorithm of visual images and detect road information under special weather conditions.This topic is a sub-topic of the major Special Project of Guizhou Science and Technology Department(Project No.Major Special Project Of Guizhou Science and Technology Cooperation ZNWLQC[2019]3012),and mainly focuses on the research of urban road information enhancement technology in downhill area under special weather environment.The details are as follows:1)The factors affecting visual visibility in foggy days and special weather at night are deeply analyzed.On this basis,the road special data set under foggy days and low illumination at night was established by using the constructed visual image acquisition platform combined with the road characteristics in mountainous areas of Guizhou.A realtime road visibility detection method based on Canny edge detection was proposed to evaluate road visibility by calculating edge intensity value.Set a certain threshold to enable image information enhancement when visibility is low.2)For urban roads in foggy mountainous areas,a video image fast fog removal algorithm based on atmospheric scattering model is proposed.Based on dark channel prior defogging,a method of dynamic calculation of atmospheric light value is proposed to improve the defogging speed and optimize the value of defogging coefficient.Moreover,CLAHE algorithm is used to further defogging and enhance image traffic information.The standard deviation and average gradient of the road image processed by the algorithm are improved by 2-4 times,the edge intensity is improved by nearly 2 times,the road visibility is improved,the real-time defogging speed is greatly improved,and the average processing time of each frame is controlled within 300 ms.3)An enhancement method based on image logarithm transformation is optimized for urban roads with low illumination at night in mountainous areas.According to the characteristics of uneven brightness distribution and low overall brightness of nighttime images,two logarithmic image enhancement methods were combined to enhance the low brightness part of the image and suppress the high brightness part.The LIP model is used to enhance the contrast by combining the features of processed images.CDF-HSD function is used to improve the image brightness.After processing,the brightness of road image is improved by nearly 4-5 times,edge intensity,standard deviation and mean gradient are improved by nearly 2 times,and the processing time of each frame is less than 250 ms.4)A classification model for vehicle detection is established by using SVM support vector machine and selecting HOG feature extractor.The model is used to classify and compare the foggy road and nighttime low illumination road data sets before and after enhancement.The results show that the accuracy,accuracy and recall rate of vehicle detection are improved by 5.04%,1.12% and 6.56% respectively.After low illumination enhancement algorithm processing,detection accuracy increased by 4.26%,accuracy increased by 3.26%,recall rate increased by 5.72%.The recall rate of both is significantly improved,indicating that the recognition ability of positive samples is improved,indicating that the research results of this paper have good application value for enhancing traffic information such as road vehicles. |
PDF Full Text Request