Research On Epipolar Geometry And Stereo Matching Algorithms Based On Underwater Fish-eye Images

There are abundant resources in the ocean.It is the key work in the field of ocean research to explore and develop the deep-sea environment with great challenges.As an important part of computer vision,binocular stereo vision is the core part of robot deep-sea exploration.However,the fish-eye lens with large view angle is of great significance for the exploration of rich marine environment and brings more abundant marine information for underwater stereo vision research.Matching technology is the core tool of vision study.The quality of the matching results will directly affect the accuracy of the 3D information of spatial points,thus determining the application value of stereo vision technology.In this paper,the underwater binocular stereo vision system is taken as the research object.Aiming at the problem that the matching accuracy of underwater fish-eye image decreases due to the double distortion and image distortion,an underwater binocular fish-eye imaging model and dense stereo matching algorithm based on spherical glass waterproof cover are proposed,which can provide some theoretical guidance for the development of underwater fish-eye vision.The main research work of this paper is as follows:First of all,aiming at the poor shooting effect of underwater plane glass mask binocular fish-eye imaging system in the real scene experiment,the underwater binocular fish-eye imaging model based on spherical glass waterproof cover is established,and the epipolar curve of underwater fish-eye image is deduced;At the same time,a gradient operator is proposed to calculate the discrete spherical image with irregular pixels,which combines the color information of the fish-eye image to calculate the cost.To a certain extent,the edge information of the image is saved and the stereo matching of the underwater fish-eye image is realized.Through the simulation experiment,the correctness of the model and the derived epipolar curve is verified.The proposed stereo matching algorithm can be better applied to underwater fish-eye images,and can get more accurate dense parallax images.Secondly,in view of the multiple distortion of underwater fish-eye images and the matching deviation caused by the poor correspondence of traditional matching window,anirregular adaptive window construction scheme based on SLIC image segmentation is proposed,which makes the edge region segmentation more accurate.At the same time,the introduction of weighted thought is used to guide the image filtering in the cost aggregation link to eliminate the influence of disadvantages.The weighted thought can better distinguish the edge and non-edge areas of the image.The results show that the proposed adaptive window construction and weighted guided filtering scheme is better than the adaptive window scheme based on Mean-Shift segmentation algorithm,which can be successfully applied to complex underwater fish-eye images and improve the matching accuracy of underwater fish-eye images.Finally,since underwater fish-eye vision does not have a test platform similar to Middlebury,nor standard test pictures and parallax pictures,this paper selects underwater fish-eye images taken in different environments for comprehensive experimental testing.In the comprehensive experiment,through quantitative and qualitative comparison with other advanced algorithms in the same environment,it can be seen that the proposed algorithm has a better matching effect in areas with large distortion and partial edge discontinuity.The average error rate of the algorithm is 5.68%.The overall experiment achieves the expected results and has a leading level in accuracy,providing reliable experimental data for underwater fish-eye vision related research.