Research On Relative Posture Detection Technology Of Large Structures Based On Multi-group Binocular Vision

With the development of modern industry,large-scale structural products represented by equipment such as satellite antennas and aircraft have increasingly higher requirements for overall accuracy in the manufacturing and assembly process.The ability to quickly and accurately detect the relative position of large structures,and ensure the assembly speed of each component of the system and the overall assembly has become a research hotspot.The detection method of binocular vision has advantages that other detection methods do not have in terms of real-time performance and degree of automation,and a single group of binocular vision system cannot perform high-precision detection of large structural products.The technology of detecting the relative pose of large structures has important research significance and practical value for the docking,welding,and assembly of large structures.This paper aims at detecting the relative poses of large structures with multiple sets of binocular vision systems.The main research work is as follows:Investigate the relative pose detection system scheme of large structures.Aiming at the problem that the size of the measured object is large and the structural features are complex and difficult to measure,according to the performance indicators,a solution for posture detection of large structures is designed.Aiming at the problem of poor real-time performance of traditional multi-eye vision image processing methods,this paper designs multiple binocular vision systems to process and transmit data.From the aspects of system reliability and response speed,the control hardware platform of binocular vision target detection system was developed.Consider the design of the package structure of the binocular vision system in this paper from the aspects of shell form,sealing and installation and positioning.The target of visual recognition in this paper is designed from the aspects of the use environment,the number of feature points,and whether the features are obvious.Determine the operating system from the aspects of reliability and efficiency and design the visual inspection process.Investigate the target pattern recognition and extraction method of binocular vision system.Aiming at the problem that the traditional Canny algorithm's edge detection is greatly affected by variance and causes discontinuities in edges,this paper introduces morphological knowledge to make the edge curve smoother.Aiming at the problem that the noise contour edges in the global image affect the processing speed,this paper sets a series of contour constraints to filter contour features.Aiming at the shortcomings of the traditional robustness method of poor robustness,an improved binary method was proposed.Aiming at the problem that the template size of the traditional template matching method must be equal to the target size in the image,which leads to the poor applicability of the template,this paper scales the image after contour filtering and segmentation to the template size,and then matches its binary image The target is segmented and extracted from the image.Aiming at the problem that the template matching method is used to match the entire image and affect the processing speed,this paper proposes a detection method that combines the template matching method with the prediction of target position information.According to the target obtained in the previous frame of the image,Position information to predict the target position of the image in the current frame and reduce the amount of calculation for image processing.The method proposed in this paper is experimentally verified.Investigate the three-dimensional positioning method of target by binocular vision system.In the study of three-dimensional target positioning method,aiming at the problem that the positioning accuracy of the target is not high due to the distortion error in the imaging process of the camera,this paper corrects the radial distortion based on the ideal imaging model and establishes a complete nonlinear camera imaging model.Aiming at the problem of excessive noise after Harris corner detection,this paper studies the rough corner selection method and introduces the calculation method of convex hull and center of gravity to obtain the two target corner points on the target pattern.Binocular vision is used.The principle of distance measurement obtains three-dimensional information of two target corners on the target.The method proposed in this paper is experimentally verified.Investigate pose detection methods for large structures.Aiming at the problem that different groups of binocular vision systems have different world coordinate systems,it is difficult to solve the relative pose of the structure,this paper proposes a method of coordinate system fusion based on the installation positions of each group of equipment.The camera must have installation errors during the installation process.In this paper,the invariance of the relative position of the binocular vision system and the attitude adjustment mechanism is used,and the camera moves a fixed distance or angle,according to the 3D data change of the detected position before the movement,Correct the target positioning error,and improve the detection accuracy.There is a certain redundancy in the solution of poses for multiple detection points,which makes complex poses of large structures.This paper introduces the singular value decomposition(SVD)method to calculate the rotation matrix through all feature points to make the relative poses between structures The solution is simplified.In order to verify the feasibility of the positioning error correction and pose calculation method in this paper,two sets of binocular vision systems were used for experimental verification under two operating conditions.