Traditional burr detection methods are increasingly unable to meet the requirements of deburring in terms of accuracy and efficiency. In this paper, machine vision is adopted. Firstly, the image of the workpiece is preprocessed, and appropriate edge detection operators are selected to detect the basic edge of the burr. Then, the closed burr edge is generated by detecting the slow changing area with the regional growth. Finally, the edge is judged to be burr by comparing the contour and other methods. Experiments show that this method has the advantages of fast, high efficiency and strong stability in the study of burr detection, and can meet the requirements of the project.

For the ellipse fitting problem often encounter in visual measurement, this paper proposes a method combining algebraic distance fitting method based on least squares and characteristic root method for ellipse fitting. Firstly, the improved hierarchical agglomerative clustering is used to denoise the data. Then the algebraic distance fitting method based on least squares is used to calculate the initial iteration value of the parameters in the characteristic root method. Finally, the Gauss-Newton iteration method is used to solve the elliptic parameters. This algorithm is used to detect the out-of-roundness of the optical fiber and compare the test results with the high-precision optical fiber tester FGM-502. The results prove the validity and accuracy of the ellipse fitting algorithm studied in this paper and show that the algorithm studied in this paper meets the requirements of actual ellipse fitting measurement.