Feature Extraction And Matching Of Multi Type Tested Target Under Complex Environment

In the process of large-scale component processing and assembly,the realization of precise measurement of relevant geometric parameters is an important link to ensure the quality of production.Computer vision measurement is a new measurement technology.Compared with the traditional measurement technology,it has the advantages of no contact,large measuring range,short time and high precision.By installing the target on the component in advance and taking pictures,the relevant measurement is completed based on the target features in the image.Due to the constraints of the complex environment such as the shooting condition and optical imaging,the tested target image often has the interference factors such as image noise,blur and object occlusion,so the accurate extraction and matching of the relevant features of the tested target image in the complex environment will help to achieve the precise measurement task of large components.In view of the influence of the complex environment on the image,this paper tries to eliminate the interference caused by the image degradation and complex background as much as possible through the image denoising,deblurring and image segmentation.The line,circle and corner features of the target are extracted by Hough transform and geometric constraints.The RANSAC ellipse feature extraction algorithm combined with edge filtering is proposed.By tracking and filtering the image edge,relatively complex ellipse features are extracted quickly and accurately.Then design the image matching algorithm based on the CenSurE-star and LDB,construct the LDB description of the CenSurE-star feature points in the image,in the image matching,take the Hamming distance as the basis and eliminate the mismatches.Combined with Structure from Motion,the matches are used to construct the 3D point cloud model of the tested target,and then the feature matching between 2D-3D of the tested target is realized.Real images are used to verify the proposed methods.The experimental results show that the RANSAC ellipse feature extraction algorithm combined with edge filtering can extract the ellipse features on the target more quickly and accurately than the general RANSAC algorithm by increasing the proportion of ellipse edges;the image matching algorithm based on the CenSurE-star and LDB is faster and ensures higher matching accuracy than the common algorithms such as SIFT.The research methods will be helpful to extract and match features of the tested target under the complex field environment with high quality.