Research And Implementation Of Real-time Mosaic Technology Of Haze Image

The purpose of image mosaic is to make wide high-resolution image,which are composed of some small view pictures with overlapping regions in the same scene,to meet the needs of observers that audience can viewing the wide range of image.Image mosaic technology is widely used in video surveillance,virtual reality,remote sensing and medical image processing.In the surveillance of large scene like the airport,docks,squares,traffic junctions,in order to obtain the global wide range of image information in real time,it is needed to set many network cameras.Although this method can obtain wider range of surveillance,the scene is divided into multiple sub-windows for display,which cannot completely and directly reflect the real scene natural.In addition,Severe fog and haze in our country makes outdoor surveillance system greatly disturbed in normal working hours.Aiming at the problem of image mosaic,this paper mainly does the following work:1)Through the research of dark channel prior image enhancement algorithm(HE)and contrast limited adaptive histogram enhancement algorithm(CLAHE),the HE has been properly improved to speed up the real-time.First,the original image is down-sampled,and a rough transmission map is estimated by the dark channel prior.Then the transmission map is refined by guided filters,the refined one is upsampled to obtain a corrected transmitt ion map.Finally,the corrected map is used to remove the haze in the image.The experimental results show that the proposed method has a great improvement in efficiency,while the haze removement effect is better than CLAHE.2)Compare and analyze the efficiency and accuracy of SURF algorithm and ORB algorithm in feature extraction and matching,we propose a method,which is based on the ORB algorithm,to compute the adaptive and optimal homography matrix.First,this paper builds the registration error function.Then,it calculates the homography transformation matrix of the current frame by the fixed frame matching method.The matrix is denoted H1.Next,H is optimal homography matrix of last key frame.We estimate the registration error using H1 and H,and select the homograph transformation matrix which with the smaller registration error as the optimal transformation matrix of the current frame and the subsequent video frame.This method solves the problem of the video mosaic error caused by the deviation of the relative position of the camera.3)In the process of image fusion,this research compared the effect of the common fusion algorithm,and proposed an improved weighted average method of fusing the overlapping region to make image smoother.