Research On Methods Of Haze Visibility Detection And Haze Removal

In recent years,fog and haze pollution has become a normal trend in China.How to manage it has become one of the focuses of today's society.The occurrence of fog and haze,as well as the problem of lower visibility,not only endangers the physical and mental health of human beings,but also brings great trouble to human activities and the travel.In particular,the visibility of the atmosphere has an important impact on the participants in traffic and transportation.Most of all,the sudden haze weather seriously endangers the driving safety of the drivers.Therefore,the detection of haze visibility has become an important part of intelligent transportation system.Specifically,aiming at the visibility detection and restoration problems causing by fog and haze weather,the following research points are put forward respectively.First of all,the existing visibility detection equipment and methods have problems of low performance-price ratio or poor applicability.How to effectively detect the atmospheric visibility is very important.To solve this problem,the paper will use the method of image processing to estimate the visibility of haze weather.Atmospheric scattering model is introduced first,and then combined with the dark channel prior and guide filter algorithm to reforce the transmission map.Finally,according to the relationship of optical visibility and the atmospheric extinction coefficient,the paper has given an estimate method for haze visibility.In this paper,we approximate the atmospheric extinction coefficient by obtaining the starting point and terminal point position information of the driveway line.A large number of experimental results have proved the accuracy of the proposed algorithm.Secondly,according to the law of highway,the model of haze image visibility grade detection based on deep learning is proposed to realize the rapid identification of the visibility level.First of all,the method creates databases in the training and testing samples by collecting a large number of road monitor redios,and uses the dark channel prior algorithm to tag the images,and then train and validite the model using a three layers and five layers convolution neural network in Caffe deep learning platform.Finally,the availavility of the model is proved by experimental contrast.Finally,considering the general application of the car camera,the paper has proposed a restoration method for haze image.This method bases on dark channel prior algorithm to get the initial transmission map.And then,to solve the high time compexity problem of soft mattiong althorithm,the paper has proposed an improved algorithm based on cross relative to total variation model.Compared with the experimental results of the existing algorithms,such as cross-bilateral filter and guide filter,this algorithm is able to obtain more accurate transmission map and better visual perception dehaze image.