Reliable Surface Extraction from Point-Clouds using Scanner-Dependent Parameters

Phase-based and time-of-flight laser scanners can be used to capture dense point- clouds of industrial plants. Our goal is to reconstruct 3D models of components in industrial plants. For robustly extracting surfaces, the standard deviation of residuals in surface fitting has large impact on the result. The standard deviation is one of basic parameters in the least-squares, robust estimate, region growing, and RANSAC. However, it is not easy to estimate the standard deviations of fitting errors in a wide range of field, because the standard deviations vary in a point-cloud, according to the sizes, distances, and materials of scanned objects. In this paper, we investigate the distributions of residuals in surface fitting using experimental data, and derive prediction functions of the standard deviations for measurement errors. Our experimental result shows that our prediction functions are effective for reliably extracting surfaces of diverse sizes, distances, and materials.