Research On Feature Extraction And Registration Methods In Stereo Vision Measurement For Large-Scale Components

Large components are the key structural parts of high-end equipment,such as aircraft and rocket,and their high-quality assembly is an important link to ensure the service performance of high-end equipment.At present,the related enterprises at home and abroad have widely used digital measurement technology to monitor the assembly quality of large components.As the core link of digital measurement,three-dimensional measurement of component surface shape can quickly obtain the point cloud of product geometry and provide data support for analyzing the deviation between the product's aerodynamic shape and the theoretical model.Vision measurement,with its advantages of high precision,high efficiency and non-contact,has been widely used in the field of aerospace digital measurement.In this paper,the method of binocular vision combined with line laser stripe and marked point auxiliary features is used to measure the surface shape of large-scale components.The methods of image feature extraction and multi view data registration in the process of field measurement are studied in deeply.The main research contents are as follows:(1)Aiming at the problem that it is difficult to extract the uneven laser stripe accurately and completely due to the complex illumination and multi-source interference in the field environment,an adaptive extraction method of the uneven laser stripe under the complex background interference is proposed.Based on the CNN network combined with the twice scan frame and the non-feature filtering criterion,the area of the laser stripe is accurately located.On this basis,the sub-regional k-means algorithm is used to segment the laser stripe features accurately.The experimental results show that this method can effectively extract the uneven laser stripe features under complex background environment.SSIM,MSE and IOU are used as evaluation indexes,and its extraction effect is better than the existing methods such as GLGM,Otsu,Graph-Cuts and FCN.(2)Considering the low accuracy of feature extraction of marked points in the complex field environment,the vision system's auxiliary light is designed,and the algorithms of feature extraction are analyzed in detail.In addition,aiming at the problem of non-public points interference with marked points matching in left and right images,a matching method based on bidirectional polar constraint is proposed.The experimental results show that this method can effectively eliminate non-public points,and achieve accurate matching of the marked points.(3)In order to improve the accuracy and stability of the registration of large-scale components,each link of measurement is analyzed.The registration scheme based on auxiliary marked points is introduced,the coordinate transformation matrix solution methods between different measuring angles are analyzed in detail,and the method of eliminating abnormal points is proposed on this basis.The factors that affect the accuracy of measurement such as camera calibration,the number and layout of public marked points are analyzed,and the corresponding error control methods are proposed.(4)The vision measurement system is built,the image processing software based on C++ /Qt is developed,and the measurement methods of laser strip feature extraction and registration are verified.The experimental results show that this method can effectively extract the features of uneven laser strip under complex environment,and the average registration error is 0.056 mm.The global measurement experiment of large composite part is carried out in the laboratory,which verifies the effectiveness of the proposed method.