Improved Ellipse Fitting Algorithm with Outlier Removal

• Published in: Ji suan ji ke xue Vol. 49; no. 4; pp. 188 - 194
• Main Authors: Guo, Si-yu, Wu, Yan-dong
• Format: Journal Article
• Language: Chinese
• Published: Chongqing Guojia Kexue Jishu Bu 01.04.2022; Editorial office of Computer Science
• Subjects: Accuracy; Algorithms; Data points; ellipse fitting|ellipse detection|least trimmed square|dual point removal|vision-based measurement; Elliptic fitting; Least squares method; Matching; Outliers (statistics); Parameter sensitivity; Random sampling; Run time (computers)
• DOI: 10.11896/jsjkx.210200040
• ISSN: 1002-137X

Outliers can significantly affect the results of ellipse fitting. Aiming at this problem, an improved ellipse fitting algorithm based on truncated least squares method and two methods based on double point removal are proposed. The truncated least squares method starts with random sampling , in each iteration, the data point with the smallest current fitting residual is selected as the fitted point set in the next iteration, and finally converges to the fitting result of the non-outlier points occupying the main body of the point set; double point removal The rule starts from the complete set of points to be fitted, and removes a pair of data points whose fitting residuals are the maximum positive and negative values ​​each time until the number of remaining points does not exceed a given ratio. On the image set of the actual part, for The proposed three algorithms and the existing comparison algorithms are tested. The results show that when the number of retained ellipse points is small, the two algorithms b