Research Of Image Visibility Detection And Defogging Algorithms For Foggy Days

Visibility detection and defogging of foggy images are the hot areas of computer vision and traffic video image processing.Visibility detection algorithm can effectively avoid the occurrence of accidents.Image defogging algorithm can expand the maximum visibility distance and further control the accident rate within the minimum range.The traditional detection method has a small detection range and a large error,and the image distortion and color oversaturation obtained by the traditional defogging method.To solve the above problems,this thesis designed the foggy image visibility detection algorithm using transmittance and scene depth and the image haze removal algorithm combined with Stacked Hourglass network model.1.Design of foggy image visibility detection algorithm based on transmittance and scene depth.Firstly,the visibility detection formula is derived according to Koschmieder’s model and the comparison threshold recommended by ICAO.Then,the extinction coefficient is obtained according to the atmospheric attenuation model,the transmittance value is obtained by dark channel prior theory,and the scene depth value is obtained by bilateral filtering combined with SFS(shape recovery from shadows)and binocular model.Finally,the visibility of the image is inverted by solving the extinction coefficient.In this thesis,the accuracy,accuracy and detection efficiency of the algorithm are greatly improved by self-made data set.2.Design of image fog removal algorithm combined with Stacked Hourglass network model.The scene depth model of SFS+ binocular model in visibility detection algorithm is used for image defogging,but the defogging effect is not ideal,and the restored image lacks edge information.Therefore,in this paper,Stacked Hourglass network model is used to optimize the depth of the image scene to achieve fog removal.First,the scene depth value for each pixel was estimated using the Stacked Hourglass network model.Then,the maximum brightness of atmospheric light is obtained by histogram weighting method,and then the color value of the image is obtained.Finally,the scene depth value,the image atmospheric brightness value,the image color value and the atmospheric scattering coefficient are substituted into the image restoration model to obtain the fog removed image.The effectiveness of the proposed method is verified by synthetic data sets and compared with similar algorithms.In the process of defogging,the problems of image distortion and supersaturation are solved,and the sharpness of image is improved.