A novel outlier-robust surface reconstruction approach for measurement dataset

Outliers pose significant challenges in surface reconstruction, as their presence in the dataset can adversely affect the robustness of the processing. In this paper, a locally progressive moving total least squares (LPMTLS) approach is introduced for the robust reconstruction of contaminated measurement datasets, wherein Student-t regression and k-medoids clustering are employed for the identification of outliers. LPMTLS detects the abnormal point in the divided support domain by progressively generating a series of references through Student-t regression, and k-medoids clustering is adopted to automatically collect the normal data as the input of next regression. After all support domains are detected, an outlier probability is defined for each point to comprehensively consider the detection results of the whole parameter domain to remove the points with high outlier probability from the dataset. The remaining points in the processed support domain are taken for local approximation. Two extreme situations including high percentage of outliers and continuous outliers are further investigated. The processing results of simulation and experiment datasets with outliers show that LPMTLS achieves high robustness to outliers.