Robust self-calibration from single image using RANSAC

Abstract

In this paper, a novel approach for the self-calibration of single image is proposed. Unlike most existing methods, we can obtain the intrinsic and extrinsic parameters based on the information of restricted image points from single image. First, we show how the vanishing point, vanishing line and foot-to-head plane homology can be used to obtain the calibration parameters and then we show our approach how to efficiently adopt RANSAC to estimate them. In addition, noise reduction is proposed to handle the measurement uncertainties of input points. Results in synthetic and real scenes are presented to evaluate the performance of the proposed method.