Calibration Problem In Two-dimensional Large Scale Precision Measurement Based On Photoelectric System

In the wave of the industrial 4.0 revolution,product intelligence is one of the core issues.In the precision measurement of products,the image detection method combined with photoelectric system has replaced manual detection,which has become a feasible solution for product intelligence.Precision measurement usually compare the standard one in the coordinate system of the object to be tested with the object.The study of converting the standard coordinate system into the coordinate system to be tested is the calibration problem that needs to be solved in this paper.The mapping between different coordinate systems can be handled by matrix transformation.According to whether the straight lines in different coordinate systems can maintain a parallel relationship,the matrix transformation is divided into rigid transformation and projection transformation.In order to establish the connection between different coordinate systems,this paper proposes a method of combining template matching and contour extraction to locate the Mark points(positioning holes)of the resonant cavity.Since the straight lines in the standard coordinate system of the cavity can maintain parallelism under the coordinate system to be tested,the rigid transformation model in the matrix transformation is used to convert the coordinates of the standard components to be tested into the current components' coordinates to complete the precise positioning of the cavity.The parameters in the rigid transformation matrix are calculated from the coordinates of the positioning holes before and after the conversion.After repeated test experiments,the accuracy and stability of the location algorithm are both good.Similarly,the mapping relationship between the stencil image coordinate system and the standard design coordinate system also needs matrix transformation.In the stencil scene,as the straight lines in different coordinate systems cannot maintain parallelism,using projection transformation completes the calibration.According to the probability distribution of the ordinate of the gravity center points of the closed graphics in the stencil,the parameter-adaptive DBSCAN clustering algorithm is adapted to divide those points with a certain density into the same cluster.The feature points are selected from the feature clusters to participate in the calculation of the projection matrix parameters,and finally the registration is completed.The experimental error data shows that the algorithm has good adaptability to the defective stencil images and most of the errors fall within ±3 pixels.For the large size stencil images,this paper proposes a different solution--based on the binary tree registration algorithm,it can cope with the computational cost brought by the large volume of data.The binary tree is used to split coordinates of gravity center points of closed graphics in different dimensions,and the segmented nodes are input as feature points to the simplified random sampling consensus algorithm,and finally the most reasonable projection matrix parameters are estimated.According to the data analysis obtained by the verification experiments,the algorithm has high error-tolerant rate and robustness,and is capable of handling the problem of large size stencil registration.