Zhang's one-dimensional calibration revisited with the heteroscedastic error-in-variables model

Abstract

Camera calibration is a necessary step to extract 3D information from 2D images. Since the 1D object is easy to construct and without self-occlusion, the 1D calibration proposed by Zhang has received many attentions. However, the progress in 1D calibration mainly focuses on reducing restrictions on the 1D object's movements. The calibration accuracy still demands improvements. In this paper, the computational model of 1D calibration is reformulated, noises in 1D calibration are analyzed with this model, and an heteroscedastic error-in-variables model-based 1D calibration algorithm is proposed. In comparison with exiting algorithms, the proposed algorithm has advantages of high accuracy with a small number of measurements, rapid convergence and weak insensitivity to initial conditions. Experiments with both synthetic and real image data validate the proposed algorithm.