Research On Composite Visual Measurement Matching Method Of Large Aviation Parts

Facing the needs of China's aviation industry development,the connecting assembly of large aviation parts has become the focus of the aviation industry because it accounts for more than 50% of the total aviation industry production.Therefore,as a key technology to inspect and guide the assembly of large aviation parts,the digital measurement of the large aviation parts' shapes and sizes has become an important research content in the aviation industry.Among the existing digital measurement methods,binocular vision measurement method has become one of the main measurement methods in the assembly site of aviation industry because of its advantages including non-contact,high precision,high efficiency and high flexibility.In order to realize the measurement of large aviation parts' surfaces,a large composite binocular vision measurement method based on auxiliary laser is proposed,which can guide the subsequent digital assembly.In the field of binocular vision measurement,image feature matching is the key to ensure the accuracy,robustness and automation of the measurement method,and it is also the research focus and difficulty in this field.In this paper,the image matching of compound binocular vision measurement system in large-scale aeronautical parts assembly site is analyzed,and the industrial image matching technology is deeply studied.The main research contents are as follows:(1)The existing binocular vision image matching methods and their principles are sorted out and analyzed.Aiming at the needs of image matching in composite binocular vision measurement,the traditional matching methods including NCC template matching,SIFT image feature matching and epipolar constraint matching are respectively utilized in auxiliary laser strip center point matching and the matching of feature markers in multi-station data mosaic.The effects are also evaluated respectively.As a result,it is demonstrated that there are many limitations of mismatching and invalid matching in the process of direct matching of industrial images captured by composite binocular vision measurement system by the three matching methods.(2)Aiming at the matching problem of laser stripe center points,a matching parameter adaptive calibration method based on non-metric distortion correction is proposed.Based on the design of a feature point layout scheme for matching parameter calibration,a non-metric image distortion correction method is introduced and a small area circular collinear point recognition and matching method in Hough space is proposed to construct the initial matching point set.High precision extraction matches 100% high accuracy one-to-one.The matching parameter matrix required for epipolar constraint matching can be acquired accurately by improved region robustness RANSAC matching parameter calculation method.Compared with the traditional matching parameter calculation methods,the accuracy is improved by more than 51%.(3)Aiming at the problem of feature point matching in industrial image for multi-station data mosaic,a new method of feature point matching based on biepipolar intersection constraint is proposed.Based on Bouguet's epipolar line correction principle,the coordinates of feature points in right image can be transformed from local perspective based on internal parameters to form a new third perspective.Then the biepipolar intersection constraint is formed by binocular images,and the searching range of corresponding matching feature points is reduced by biepipolar intersection to achieve the matching of feature points.At last,the matching accuracy reaches more than 95%.(4)On the basis of the above research,the matching accuracy verification experiments of the plane standard sample in the laboratory and the large aviation part in assembly field were carried out respectively.The results show that the plane reconstruction accuracy RMSE is0.1693 mm in laboratory environment,and the matching accuracy using feature points represented by spherical visual targets in field environment is 100%.At last,the multi-station data mosaic effect of the large aviation part is good and meets the accuracy requirements of aviation field measurement.