An efficient approach for symmetry detection in point clouds of constructions

Abstract

The symmetry in constructions is fundamental and universal across eras, continents, and cultures. Recent advances in sensing technology, such as photogrammetry, airborne and mobile laser scanning, and augmented reality, gave rise to boomingly available and affordable 3D point clouds of constructions and cities. However, detecting the symmetries in the point clouds remains very challenging, which indirectly turns many building information models (BIMs) inaccurate in geometry, massive in size, impoverished in semantics, and unaesthetic. This seminar will first revisit the fundamentals of symmetry and symmetry detection. The problem of symmetry detection is then formulated into a nonlinear optimization model and further approximated with the octree-based feature points. The computational experiments on nine real cases evaluated and confirmed the performance of the proposed approach applying state-of-the-art derivative-free optimization (DFO) algorithms. For example, the proposed approach accurately detected the global symmetries in a few seconds for 1.4 million points, which was significantly faster and even more accurate than the conventional pairwise voting-clustering methods. Research opportunities and possible collaborations will also be discussed.