| In recent years,frequent haze and other bad weather have a serious impact on the normal work of outdoor video surveillance system.In the haze weather,video or images captured through the camera often appear color shift,contrast reduction,the overall grayness and other image degradation problems.This greatly restricts the effectiveness of the video surveillance system to play and brings trouble to the safe social construction,intelligent traffic management,urban information construction and other social engineering.Currently,the study of the single image defogging algorithm represented by the dark channel a prior has made great progress,but there are relatively few researches focusing on video defogging algorithms.The frame by frame video defogging algorithm based on the dark channel prior does not consider the video characteristics,and a serious time redundancy on the transmission estimation is aroused.Therefore,it is necessary to improve the frame-by-frame video defogging algorithm by exploiting inter-frame correlation.In addition,the performance of video defogging algorithm is largely dependent on the performance of single image defogging algorithm.In order to meet the real-time application of the project,a relatively simple and fast single image defogging algorithm differing from the dark channel prior is studied,which is expected to meet the real-time requirement for monitoring video defogging occasions by algorithm improvement and optimization of system implementation.This work has a strong practical value.In view of the above problems,this article starts a study of defogging surveillance video from the following aspects:1)Based on the analysis of video characteristics,a video defogging algorithm based on inter-frame block matching is proposed using the theory of dark channel prior.Similar to the video coding,in the defogging process,the fog sequence is firstly divided into groups,which currently contains I,P frame only.Rough transmission of I frame is computed by dark channel prior.Rough transmission of P frame is obtained by inter-frame block matching.Both of them then will be optimized and corrected by guided filtering to obtain the final transmission.For proposed algorithm,the transmission is calculated once in a group,which avoiding some unnecessary calculation during the process of transmission estimation of video sequences through frame by frame method and improve the integrity of the defogged sequence.2)In order to meet the real-time requirement of video processing,given that the efficiency of the single image defogging algorithm has a great influence on the efficiency of the whole video defogging algorithm.This paper proposed an improve approach based on an image defogging algorithm which estimates the transmission point-by-point using the approximate linear relationship between fog and clear image.An adaptive parameter is employed in the improved method to estimate transmission and could handle the sky area properly,as well as the bright objects in near-field.Through the matlab simulation,the improved algorithm is about 2.5 times faster than the dark channel defogging algorithm,and because of the point-by-point estimation of the transmittance method,the data coupling between the transmission estimation is avoided,which brings the benefits to the real-time implementation.3)The improved single image defogging algorithm in 2)is extended to surveillance video defogging algorithm by using the inter-frame difference framework.Then it is implemented on the Visual Studio platform of ordinary PC using c and Opencv programming.Experimental results show that in the ordinary PC,the proposed algorithm achieved 46 fps dealing with 640*480 resolution fog monitoring video,which can meet the real-time processing requirement.Compared with the existing fast video defogging algorithm with the same platform,the proposed algorithm could not only improve the detail of the image,but also have a more fast speed. |
PDF Full Text Request