Underwater Target 3D Reconstruction Based On Trinocular Vision

In underwater exploration,underwater robots need to explore the underwater environment with cameras;Repairing the damaged part of the ship underwater requires the underwater robot to inspect the damaged part and repair it;A large amount of industrial wastewater will be produced in industrial production.There are parts scattered at the bottom of the wastewater tank.To obtain these parts,underwater robots need manipulators to salvage materials.Due to the common problems of color deviation and uneven illumination in underwater images,it is necessary to provide an efficient underwater vision system for underwater robots.Therefore,the research on 3D reconstruction of underwater targets has important practical significance.The main contents of this paper are as follows:Firstly,the camera parameter calibration involved in 3D reconstruction is studied,the transformation relations of four common coordinate systems are analyzed,different camera calibration algorithms are classified,and the scenes used by different calibration algorithms are introduced.The chessboard calibration method is used as the camera calibration algorithm in this paper.Through the underwater part image acquisition,and building the underwater part image set,aiming at the problem that there is light refraction in the image acquisition process,resulting in the underwater target is larger than that in the air,the underwater refraction compensation model is innovatively used,through the modeling of camera calibration parameters,glass thickness and glass refractive index,and then the target is compensated.Experiments show that the refraction compensation model can restore the original size of underwater target.Secondly,aiming at the common problems of strong color deviation,low contrast and uneven illumination in underwater images,an underwater illumination uneven enhancement algorithm based on image fusion is proposed.The algorithm is realized by fusing a variety of image algorithms with multi-scale and different weights,using standard data sets UIEB、Ocean Dark and underwater parts data set that I built experiment on multi scene underwater images.Qualitative experiments show that the algorithm corrects the problems of uneven illumination,poor contrast and color deviation;Quantitative experiments show that the algorithm has good performance in average gradient,underwater image quality standard and comprehensive image evaluation index,which reflects the advantages of the algorithm.Then,aiming at the problem that the traditional census cost matching is easy to be affected by the central pixel,the central pixel is modified by the uniform value of gray distribution,and a stereo matching method based on fusion cost function is proposed.The algorithm is realized by fusing a variety of cost functions and using segmentation tree cost aggregation.Experimental results show that the algorithm performs better than the traditional algorithm in the evaluation index of error matching rate.The three-dimensional coordinate solution of the target object under the parallel and non parallel binocular vision model is deduced.Aiming at the problem of limited binocular vision field,the collinear binocular vision model is used,and a parallax fusion algorithm is proposed.By fusing the parallax images obtained by different camera groups,the final parallax image can be improved.Finally,the underwater target 3D reconstruction system is built,and the 3D reconstruction experiment and size measurement of underwater parts are carried out.Experiments show that the system has high dimensional measurement accuracy and good three-dimensional reconstruction effect.