Research On Vision-based High-precision Real-time Positioning Guidance Technology

The vast ocean area and abundant resources are the focus of future development.The Eighteenth National Congress clearly put forward the goal of "building a maritime power and vigorously strengthening naval construction." The 19 th National Congress of the People's Republic of China further requested the implementation of the marine strategy,and discussed the construction plan of the modern marine development system.With the continuous advancement of marine economic development and the "Deep Blue" plan,long-term offshore operations have become the norm.One of the most critical issues facing fleets or ocean-going platforms away from shore is the supply of materials.In view of the problem that two ships rely on supplementary materials and collisions are likely to occur when the container is lifted,this paper proposes the use of multiple sets of binocular cameras combined with manual marking points to detect the six-degree-of-freedom posture of the container,and the detection results are used for subsequent motion compensation solution.To this end,the following studies were conducted:Firstly,The basic theories related to binocular camera imaging and calibration are introduced.Starting from the linear and non-linear model of the camera pinhole imaging,and the conversion relationship between the four coordinate systems involved in the imaging process,the general mathematical model of the binocular camera is derived;the principle of spatial coordinate measurement of the binocular camera and possible The source of the error and the method to reduce the error;the camera calibration method and principle are briefly described,and the calibration results of different calibration methods are compared through specific experiments.Secondly,The feature point extraction and tracking methods based on Ar Uco artificial marker points are analyzed in detail.Introduced the common image preprocessing method and the process of fast,stable and high-precision feature point extraction in the binocular vision system;introduced some existing image feature extraction algorithms and principles.The three main indicators of accuracy,time-consuming and robustness of each algorithm,select the optimal feature extraction algorithm suitable for this article;in view of the real-time requirements,the artificial marker Ar Uco Tag is introduced to assist in feature extraction,and the Ar Uco marker point corner is introduced The method of extraction and tracking,and verified the rationality of the method through experiments.Then,the method of stereo matching of the binocular camera and the solution of the target pose are described.The types and main principles of mainstream binocular stereo matching algorithms are introduced,and a point feature stereo matching method based on epipolar constraints is proposed;using the extracted feature point data,the three-dimensional space of Ar Uco artificially marked corners is obtained through stereo matching Coordinates and use the SVD decomposition method to decompose the rotation and translation matrices R and T to calculate the target pose;the feasibility of the method is verified through experiments.Finally,according to the actual container stacking process,a platform was built and different experiments were designed to test the accuracy,speed and robustness of visual pose recognition under different lighting conditions.The test results show that: the average error of rotation angle measurement of the binocular vision system is ±1.1°;the average error of displacement measurement is ± 0.4%;and the minimum recognition speed is above 15 Hz.Can meet and rely on the actual requirements of hoisting.