Research On High Frequency Transformer Size Visual Measurement Method Based On Multi-feature Invariant Fusion

This paper focuses on the measurement of high-frequency transformer size,using binocular vision measurement technology to measure its size.Firstly,the binocular vision three-dimensional measurement system is introduced.The measurement process of the binocular vision measurement system is analyzed.The hardware structure and software algorithm of the measurement system are briefly introduced.Secondly,because the binocular camera is used,the binocular camera needs to be calibrated.Zhang's calibration method is studied and improved.After the calibration is completed,the high-frequency transformer picture is acquired,the invariant features are extracted and merged,and the fusion feature is stereo-matched using the region feature matching method to obtain the parallax image,and then according to the parallax image.3D information recovery and dimensional measurement.The main research results are as follows:(1)Studying the Zhang's calibration method and improving the corner point extraction algorithm in the checkerboard.The corner extraction result is compared with other corner detection methods.The results show that the corner coordinates of the checkerboard can be completely detected in a complex environment,while others The algorithm detects corners in the environment.The improved algorithm detects the coordinates of the corner points in the checkerboard to the sub-pixels and substitutes them into the Zhang's calibration formula.Finally,the binocular camera parameters are obtained,and the cumulative error of the corner points extracted in the checkerboard is calculated to be 0.31 pix.(2)For the SIFT operator to generate feature points,the time-consuming shortcoming is improved.The SIFT feature extraction algorithm is improved.The feature point description of128 dimensions is simplified to 16-dimensional by double-ring template.The effectiveness of the improved algorithm is verified by experiments and SURF.The operator is compared with the original SIFT operator feature point detection and matching results.In this paper,the accuracy of feature point matching can reach 97.5%,which is significantly higher than the other two algorithms.The matching algorithm is shorter than the original SIFT method and slightly longer than the SURF algorithm,but it does not affect the improvement of this paper.The effectiveness of the SIFT feature extraction algorithm.(3)Combining the Harris corner feature with the improved SIFT feature and combining the edge feature and the line feature can accurately find the binocular image matching feature points at both ends of the high-frequency transformer pin,restore the three-dimensional spaceand calculate the stitch size.The calculated stitch size results are compared with the vernier caliper measurement results and the two-dimensional image size measurement results.The results show that the binocular visual size measurement is similar to the actual one,and the standard deviation is 0.0197 cm,which is obviously better than the standard deviation of the two-dimensional image size measurement of 0.0325 cm.The three-dimensional size calculation result is more accurate than the two-dimensional image size measurement result,and the error is small.