Kluger Florian Ackermann Hanno Yang Michael Ying Rosenhahn Bodo

Abstract

We present a novel approach for vanishing point detection from uncalibratedmonocular images. In contrast to state-of-the-art, we make no a prioriassumptions about the observed scene. Our method is based on a convolutionalneural network (CNN) which does not use natural images, but a Gaussian sphererepresentation arising from an inverse gnomonic projection of lines detected inan image. This allows us to rely on synthetic data for training, eliminatingthe need for labelled images. Our method achieves competitive performance onthree horizon estimation benchmark datasets. We further highlight someadditional use cases for which our vanishing point detection algorithm can beused.