Vanishing Points Detection and Line Grouping for Complex Building Façade Identification

Abstract

We describe a new method for automatically detecting vanishing points and group lines associated with building facades from a single uncalibrated image. Accurate vanishing points detection for buildings is of importance for building façade rectification and 3D scene reconstruction. The challenges arise from confusing scene clutter, occlusions, unusual building shapes and non-Manhattan street layout. Buildings usually possess many straightline features, e.g., window and door openings, as well as their overall outline. We exploited these features and propose a robust line grouping technique. The method was evaluated on images from the Zubud-Zurich building database. The experiment shows the proposed method works well with varying building structures and image conditions, as well as filtering out “non-building” vanishing points (e.g. vanishing directions detected from road boundaries).