Study On Visibility Detection Of Traffic Vedio And Image Dehazing In Foggy Weather

Video and image processing in foggy weather has attracted increasing attention of researchers.Due to lack of light in foggy days,videos we obtained have degeneracy phenomena with low contrast,color distortion,and blurring details.This can not only bring difficulty to the computer processing and judgment,but also cause great risk to traffic safety because of unclear visibility.So,effectively improving the quality of degraded image and real-timely detecting visibility information are urgent need.It has theoretical and practical significance to reduce the impact of bad weather on life and travel.This paper combines with several national and provincial projects.Such as "Human Visual Representation of the Video and Traffic Visibility Detection"(No.61105015),"Traffic Visibility Detection Based on Human Visual Reconstruction"(No.BE2011511),and“Study on Key Techniques of Traffic Disaster Weather based on Video Sensor and Emergency Early Warning System"(No.BK2011747).Deep Researches are carried out on two respects:visibility detection of traffic video and foggy image restoration.After the study of atmospheric scattering theory,this paper describes the incident light attenuation model and atmospheric imaging model.Then a visibility detection algorithm based on optimization of squared differences with apparent luminance of roads is put forward,which is compared with current advanced video detection methods.This paper also proposes a nighttime visibility detection model on the basis of atmospheric point spread function called APSF.After that,two types of image restoration methods are introduced and validated,which are based on image enhancement and physical models.Integrating quad-tree subdivision,dark channel prior,guided joint bilateral filter with logarithmic image processing framework for color images,this paper presents a new defogging algorithm which matches human visual characteristics.An acceleration method with guided filtering is proposed afterwards,so that it reaches a significant result.The paper's characteristics and innovations are summarized as follows:?Optical reasons for the formation of fog and image degradation model are specifically elaborated.Primarily the role between atmospheric particles and light is briefly described.According to the optical absorption principle,this paper analyses the model of the incident light attenuation and the atmospheric optical imaging model,which are bases of this paper.?A visibility detection algorithm based on optimization of squared differences with apparent luminance of roads is put forward innovatively.According to Koschmieder theory,the objective function of extinction coefficient optimization is established by the least square between calculated values and reference values of apparent luminance.With experiments,this algorithm shows high accuracy,fast speed and fine robustness.?A nighttime visibility detection model on the basis of atmospheric point spread function(APSF)is brought forward.Assuming that the weather conditions are known and stable,the halo distribution of fixed point light source image at night can be analysed by APSF.Then the optical thickness and corresponding nighttime visibility are easily solved..Two major types of image dehazing methods,which are based on image enhancement and physical models,are inductively discussed.Typical Methods,like histogram,Retinex theory,Fattal's,Tarel's,dark-channel estimation,and He,s guide filtering algorithm,are introduced in detail This paper also verifys their accuracies and compares their effects.?A new defogging algorithm conformed to human visual characteristics is proposed and achieved.Firstly an adaptive method to calculate the atmospheric light on the basis of quad-tree subdivision is given.Dark channel prior is combined with guided joint bilateral filter to estimate more accurate transmissivity.Then LIPC is used to further enhance the result.Guided filtering is adopted to do upsampling in order to accelerate the algorithm.Eventually the algorithm is verified reliable and effective by subjective comparisons and objective analyses.