The Research On Dense Stereo Matching Algorithm Of Fish-eye Images

With the continuous development of artificial intelligence,computer vision technology has been more and more widely used,lens is an integral part of the computer vision system.Ordinary lens' viewing angle is usually 50?-60?,narrow viewing angle has been unable to meet the needs of many occasions.But fisheye lens with large field of view,can take a rich scene information one-time,this advantage makes it attracte more and more attention in the civilian,industrial,military areas.Among them,fisheye lens binocular stereoscopic vision technology is one of the hotspots in computer vision research,and has great prospects in large-scale three-dimensional reconstruction and robot positioning and navigation.This paper systematically studies the modeling construction,the polar line,the window construction and the matching algorithm of the binocular fish-eye cameras in the air and underwater respectively.It is of theoretical significance to the application of the fish eye binocular stereo vision.The main contribution as follows:Firstly,from the stereo matching frame of ordinary perspective image,the basic theory of the pole geometry,the point of the same point and the parallax theory in stereo matching are introduced.The existing stereo matching algorithms are systematically classified and analyzed.The general algorithm of regional matching algorithm is briefly summarized,which is the foundation for the region matching algorithm of fish-eye image.Secondly,aiming at the fish-eye images in the air,a dense matching algorithm based on Adaptive Support-Weight approach for fish-eye image is proposed.According to the fish-eye image of the imaging model,the polar curve of fish-eye image is deduced and the irregular window adapted to the fish-eye images is established.In order to take into account the speed and accuracy,the initial disparity map is obtained by using the Adaptive Support-Weight approach whose windows are based on segmentation and smaller.Then,the initial disparity map is corrected to obtain an accurate disparity map,the adaptive weights are redefined according to the characteristics of fish-eye images in the correction algorithm.The algorithm is validated by the simulated image and the real imagerespectively.The algorithm is applied to the fish-eye image in the air successfully.Thirdly,aiming at the underwater fish-eye images,the underwater binocular fish-eye camera system model is established and the polar curve of the underwater image is deduced.In order to make the subsequent matching algorithm more accurate,an adaptive window based on color segmentation is proposed and the whole stereo matching of the underwater fish-eye images is completed.The experimental results verify the correctness of the model and the derived polar line,and the constructed adaptive window can be better applied to the stereo matching algorithm of the underwater fish eye image.