Research On 3D Reconstruction Of Underwater Surroundings Based On Parallel Binocular Vision

In recent years,research on 3D reconstruction based on parallel binocular vision is an important research element in the mobile robot field.The fundamental principle of binocular vision has a strong resemblance to the visual perception procedure of humans.We can get different images with two different view points from two different parallel cameras,then calculate the disparity of images pixels based on triangular measure.Finally we can reconstruct the 3D coordinate of the object.Then the surface of objects can be reconstructed using this depth information.One complete 3D reconstruction system based on parallel binocular vision contents six parts: image collection,camera calibration,image feature extraction,image matching,the calculation of depth information and 3D reconstrucion of the surface of objects.The camera calibration and image matching are very important among them.In this paper, the traditional theory of binocular vision is studied. Then combining with the practical requirements of 3D reconstrucion of the underwater surroundings for the AUV(autonomous underwater vehicle), a research on the imaging models of the parallel binocular vision and calibration of camera parameters is carried out. Then the paper designs experiments to achieve the practical requirements. In the later chapters, extraction of the feature points and image matching, generation of disparity map, the influence of refraction, as well as the calculation and display of three-dimensional information are all studied. In the same time, experiments have been done to verify the effectiveness of the method in all parts. The main works and innovation of the paper are as follows:(1) Zhang Zhengyou camera calibration method is studied in the first place. Then through the design of experiments, the paper calculates the key parameters of the cameras. Many problems that should be paid attention are also given appropriate solutions.(2) Image matching based on point feature is presented in this paper. Firstly, the Harris corner detector is used to detect corners.Then the keypoint descriptor of SIFT(Scale Invariant Feature Transform) is introduced to describe the direction character of the Harris corners. Secondly, Euclidean distance is performed to get exact matchings in this method. Finally, an easy and effective method to avoid wrong matches is discussed. This method is effective on brightness invariant, movement invariant and rotation invariant in the image matching of the binocular vision. It is proved that this method can be used in practice.(3) One new mode of the disparity map is introduced.Considering the influence of refraciton,a new way of 3D reconstruction based on disparity map for the underwater surroundings is proposed.We finally use the OpenGL to display the 3D points,which can meet the requirements of the project.