Residual uncertainty in three-dimensional reconstruction using two-planes calibration and stereo methods

摘要

A necessary goal in the calibration of any instrument is an indication of how well the particular device performs. In the specific case where a camera system is used in the process of three-dimensional reconstruction or stereopsis, geometric measures of the accuracy, precision, and repeatability of 3-D point measurements give a quantitative evaluation of the success of that algorithm. This paper presents an experimental performance analysis of a traditional calibration algorithm as applied to the problem of three-dimensional vision. The study begins with an analysis of the algorithm in a single camera or monocular system. Monocular analysis concludes with a derivation of error boundaries under different geometric constraints. Showing how this residual error propagates from the component monocular systems into a three-dimensional reconstruction, a binocular analysis derives the desired performance monitors. The paper concludes with a geometric interpretation of the results.