Research On Methods For Haze Visibility Detection Based On Road Surveillance Videos

The emergence of haze pollution causes great losses to society,and by the community’s close attention.Haze pollution leads to reducing visibility and seriously affects people’s outdoor activities and traffic travel.In particular,sudden mass fog weather brings a huge accident hidden trouble to the traffic.It is well known that accurate visibility detection is one of the necessary steps of solving the problem,but the accuracy and applicability of existing methods need to be improved urgently.Therefore,focusing on people’s health and travel,it is urgent to find a real-time and effective haze visibility detection method.Haze visibility estimation has become a hot topic in the areas of image processing and computer vision and attracted lots of researchers’ attentions because of its theoretical and practical values.In order to overcome various disadvantages of traditional visibility estimation methods,researchers study the visibility estimation based on the camera calibration technique,image edge detection,and machine learning means.In this paper,the principle of visibility detection is analyzed,and the road visibility is estimated by using surveillance video images.The main contents of this thesis are included as follows:First of all,in view of the shortcomings of estimating transmittance based on dark channel prior theory,the transmittance is optimized by a parameter correction method.This paper constructs a function to describe the brightness of the sky based on the pixel-based bright channel prior theory,which reduces the error when the sky brightness is acquired.In addition,the guided filter is used to eliminate the blocking and halo effect.Last but not least,a fast lane-line detection method is designed to help to estimate the visibility.Lots of experiments are performed to demonstrate the validity of the method.After demonstrating the feasibility of using image entropy to detect haze visibility,an algorithm of highway visibility detection based on minimum image entropy is proposed.In this paper,the road region of haze image is extracted to calculate the dark channel and atmospheric transmittance based on a region growing algorithm.Than the scene depth information of the road region can be calculated by using the lane prior information.By retrieving the extinction coefficient,the recovered image is obtained with the atmospheric scattering model and the image entropy of road region can be calculated.Finally,the atmospheric visibility of the haze image can be obtained by searching the extinction coefficient corresponding to the minimum value of the image entropy.The experimental results show that the algorithm is consistent with the human observation,which meets the safety requirements of highway.