Quick, unconstrained, approximate l-shape method

Abstract

This simple paper describes an intuitive data-driven approach to reconstructing architectural building-footprints from structured or unstructured 2D pointsets. The function is fast, accurate and unconstrained. Further unlike the prevalent L-Shape detectors predicated on a shape's skeletal descriptor [Szeliski 2010], the method is robust to sensing noise at the boundary of a 2D pointset.