Research On The Calibration Method Of Binocular Stereo 3D Measurement System

Binocular stereo 3D measurement technology has important relevance and a wide application background in the fields of industrial measurement,aerospace and heritage restoration because of its non-contact,high accuracy and flexibility.In order to achieve accurate measurement of objects,it is necessary to firstly complete the calibration of the parameters of the two cameras separately,and secondly to determine the positional relationship between the left and right cameras,i.e.to complete the system calibration work.Based on the system calibration results,the 3D data of the feature points can be obtained,thus achieving accurate measurement.In the whole process,the calibration accuracy of the binocular stereo vision has a very important role in the subsequent measurement results.In this paper,the accurate calculation and system calibration of center image of circular array target in binocular three-dimensional measurement system are studied.The main research contents and results are as follows.(1)The problem of locating the central image point of a circular array target: In the field of computer vision,circular targets are widely used in all kinds of complex visual scenarios due to their obvious geometric characteristics,and the center of the circle is usually its important feature point.However,in practical measurement,the target plane and the camera imaging plane are usually not exactly parallel,the circular marker point is projected in perspective and then receded into an ellipse,and the geometric center of the ellipse is not the true projection point of the center of the circle.In order to improve the accuracy of central image of circular array target,a fast method of extracting central image of circular array target using geometric features is proposed.Firstly,the acquired target image is subjected to edge detection,shape filtering and sub-pixel edge extraction,and the general equation is obtained by fitting the ellipse;secondly,the coordinates of the two vanishing points in the direction of the target coordinate axis are calculated through the projective transformation matrix;finally,the coordinates of the two sets of tangent points of the ellipse are solved by associating the vanishing points with the general equation of the ellipse using the principle of perspective invariance,and the intersection of the lines connecting the two sets of tangent points is the projection of the circle center of the circular marker point on the image plane.The intersection of the two sets of tangents is the projection of the center of the circle on the image plane.The intersection of the two sets of tangent points is the projection of the circle on the image plane.The experimental results show that the proposed method can improve the accuracy and simplify the solving process.(2)System calibration issues:System calibration is a fundamental task in the field of vision measurement and 3D reconstruction,and the accuracy of its calibration parameters determines the quality of subsequent measurement results.Therefore,a method of binocular stereo calibration based on data filtering is proposed.The factors affecting the calibration accuracy of the system can be generally divided into four aspects: the accuracy of extracting feature points on the captured target image,the calibration algorithm used,the accuracy of the target marker point’s own 3D coordinates,and the pose of the target during the binocular system.The accuracy of the target marker’s own 3D coordinates is determined by the production process and is generally not considered.For the factor of accuracy of extracting feature points on the shooting target image,the accurate extraction of feature points of flat target is realized based on the principle of perspective invariance.For the factor of calibration algorithm used,Zhang Zhengyou’s calibration method is used to realize the calibration work with higher accuracy.For the pose problem of target images taken by binocular system,the target images are acquired based on the multi-azimuth set calibration method,and the threshold value is set to filter the acquired target images and filter out the positional images that affect the calibration accuracy.the final experimental results show that the proposed method can improve the calibration accuracy as well as the measurement accuracy,and can meet the high accuracy requirements in the current industrial production measurement field.