| With the continuous development and utilization of marine resources,underwater robot technology has been rapidly developed and underwater visual system has been developed too.Due to the features of the vision system such as: the non-contact measurement,large amount of information,low cost and so on,which has become an important part of underwater equipment.Underwater 3D reconstruction technology is widely used in underwater autonomous aircraft operations,seabed exploration,pipeline maintenance,underwater target recognition and other fields,so the development of underwater 3D reconstruction technology has important academic significance and practical value.The purpose of this paper is to apply the binocular vision system to underwater,achieving 3D reconstruction of underwater scene,providing the location information of jab target for an underwater robot.Binocular vision is to imitate human eyes to observe the same scene,which combined with the camera's imaging principle,observing the same object with binocular and obtaining the parallax,restoring the depth information of a three-dimensional space object by the principle of triangulation.Usually,a complete binocular vision 3D reconstruction system is composed of six parts: image acquisition,camera calibration,image feature extraction,stereo matching,depth information calculation and 3D reconstruction of object surface.Firstly,this paper start with the imaging model of the camera,creating four coordinate systems,and given the conversion relationship between the coordinate system.Then,the camera calibration method is classified and the applicable scene of each calibration method is introduced,finally,the calibration method of Zhang Zhengyou is given,the calculation method of the internal and external parameters of the camera is derived.Secondly,the features and difficulties of underwater image are analyzed.In the processing of image enhancement,three kinds of smoothing methods are tried,and through experimental comparison,this paper chooses the best way to handle.The contrast is enhanced by using two sharpening methods and histogram equalization,and an improved fogging algorithm combining homomorphic filtering and histogram equalization is proposed.In the processing of image restoration,a image degradation model is established,and three filtering methods are used.Besides some improvements based on the constrained least-squares filter have been made,and the restoration effect of the image has achieved an ideal effect.Thirdly,the Harris corner detection and FAST feature point detection method are explained,and in the Harris corner detection theory,the method of curve fitting in X and Y directions is introduced,which took the peak point as a corner point.As well as experiments show that this method is superior to the traditional corner detection method;In the stereo matching processing,the classification of matching algorithms and the key technology of stereo matching are introduced in detail,four matching rules are summarized,and the principle of outer limit constraint is introduced emphatically,then,the concept of similarity measure is given.In stereo matching,a global stereo matching method named dynamic programming stereo matching method is proposed,and the depth image is obtained by the verification.Finally,the 3D coordinate calculation method of target objects under both parallel and non-parallel models is derived.on the OPENGL platform,triangulation method is used to reconstruct the object surface,then the size of the object is calculated based on these data.Besides through the test comparison,the target precision is reached and the construction of underwater 3D reconstruction system is realized. |
