A novel reconstruction method with robustness for polluted measurement dataset

Due to its capacity to depict intricate geometric shapes and topological structures, moving total least squares (MTLS) method has garnered considerable attention from a diverse spectrum of researchers, finding extensive utility in the domain of reverse engineering. Nonetheless, the utilization of MTLS becomes impractical when dealing with datasets containing outliers. Drawing inspiration from the construction principle of MTLS, a regionally asymptotic moving total least squares (RAMTLS) approach is proposed to achieve robust reconstruction of highly polluted measurement dataset, in which the Bayesian estimation and density-based spatial clustering of applications with noise (DBSCAN) are used for outlier detection. In contrast to the conventional method, the proposed approach employs a step-by-step iterative strategy to mitigate the adverse impact of outliers. Within this framework, Student-t distribution-based estimation with non-prior information is used to pre-fit the data within the support domain, followed by clustering the acquired absolute residuals. Upon the completion of the regression-clustering iteration, a weighted Bayesian estimation is further applied to the remaining data to ascertain the ultimate estimated value. Compared with existing competitive methods, the simulations and experiments emphasize the effectiveness and reliability of the proposed reconstruction method.