Line fitting in Euclidean 3D space

Abstract

Over a century ago Pearson solved the problem of fitting lines in 2D space to points with noisy coordinates in both dimensions. Surprisingly, however, the case of fitting lines in 3D space has seen little attention, though Adcock long ago published a brief (one page) article claiming that the solution that minimized orthogonal distances is the most probable. We solve this problem using a new algorithm for the Total Least-Squares (TLS) solution within an Errors-In-Variables Model, respectively an equivalent nonlinear Gauss-Helmert Model. Following Roberts, only four parameters are estimated, thereby avoiding over-parametrization that may lead to unnecessary singularities and, hence, require the introduction of constraints to the model. The current pervasiveness of Global Navigation Satellite Systems, robotic total stations, and digital laser scanners as sources of geodetic observations means that geodetic engineers and scientists now commonly work with observational models in 3D space as opposed to classical geodetic methods that often separated horizontal and vertical observational models. And while several papers have been written describing a TLS solution for line fitting problems in 2D space, the extension to 3D space is not readily apparent from these works. This further motivates the treatment of the 3D problem in some detail in this contribution.