Study On Feature Extraction Methods Based On Stereo Vision Measurement For Large-Scale Aviation Components

In the process of production,manufacturing and transportation,large scale aviation parts inevitably have some dimension error and shape deformation.The error accumulation of different links leads to the difficulty of assembly coordination control.At present,foreign aviation manufacturers have basically realized the integration of digital design,manufacture and assembly.One of the key steps is the digital measurement of parts.By using the digital measurement methods,the actual shape information and geometric dimension of the parts are restored and compared with the theoretical model to guide the completion of high quality aviation parts assembly.Binocular stereo vision measurement has been widely used in the process of aviation digital assembly because of its high measurement accuracy,good robustness,high efficiency and wide measurement range.In this paper,a binocular stereo vision measurement method based on aided laser scanning is used to realize the three-dimensional measurement of large aviation components.The research of image processing methods such as laser stripe feature extraction and boundary feature extraction are carried out.The main research content is as follows:(1)Aiming at the problem that the intensity distribution of the laser stripe is not uniform,and the laser stripe feature is difficult to be extracted quickly and accurately,a laser stripe center extraction method based on self-adaptive threshold processing is proposed.An adaptive threshold mathematical model of laser stripe region segmentation is established.On the basis of this,a local quadratic curve approximation method is used to quickly obtain the normal direction of the laser stripe and then extract the precise laser stripe centers.The experimental results show that compared with the Steger algorithm,the center extraction error of laser stripe is less than 0.1pixel,and the extraction efficiency is increased by about 3 times.(2)Aiming at the problem of the complex on-site illumination environment and the serious background interference,which always cause that the boundary of the measured part is difficult to be detected completely and accurately,a boundary extraction method based on feature mutation of laser stripe is proposed.The geometrical mutation characteristics of laser stripe at the target boundary region are analyzed firstly.Then the ideal boundary model is established.A combination of rough and accurate extraction is utilized to extract the boundary feature points with high precision.The experimental results show that the proposed method can effectively extract the object’s boundary feature under complex background environment.The precision can reach 0.056%.(3)To further improve the efficiency of image processing,an improved Kalman filter algorithm is adopted to fast track and locate the region of interest(ROI)of laser stripes in the sequence images.Then the laser stripe center and boundary feature can be quickly extracted in the laser stripe ROI.The experimental results show that the laser stripe ROI can be effectively tracked and the image processing in the ROI can improve the efficiency by more than 62%.(4)The binocular stereo vision measurement system based on aided laser scanning is built.Then the proposed laser stripe center extraction method and boundary extraction method are verified by using the standard part in the laboratory.The experimental results show that three-dimensional reconstruction error is less than 0.06 mm.A field experiment of a large fuselage panel part is carried out in the aircraft manufacturing factory,the feasibility of on-site application of the proposed methods is verified.