A fast knowledge-based plane reconstruction method from noisy 3D point cloud data

Abstract

This paper presents a knowledge-based method of planar surface reconstruction from noisy 3D point cloud data which produces a fast and reliable modelling of an indoor environment. The data used is obtained from a laser scanner, which is known for its rapidness and accuracy in 3D reconstruction. The laser is attached together with a servo motor at a 45cm height mobile platform to give a complete 180° from top to bottom, and 180° from left to right, which allows the reconstruction process to be made on just a single scan. It takes about 1.3 seconds for the algorithm to process and produce the 3D modelling of all existing surfaces on a normal working computer. Our method has been tested in both uncluttered and real office environment. It can be applied towards developing an as-built building information models (BIM) in architectural and semantic mapping for robotics applications.