Rapid Image-based Localization using Clustered 3D Point Cloud Models with Geo-Location Data for AEC/FM Mobile Augmented Reality Applications

摘要

In this paper, we present a new method for supporting onsite construction and facility management tasks by allowing field personnel to automatically have access to the latest project information in form of Augmented Reality (AR) overlays, visually document onsite issues/progress, and communicate information with other personnel on or off site. Our near real-time and marker-less mobile augmented reality solution builds on top of a new image-based localization method for 3D point clouds that have been reconstructed using a Structure-from-Motion (SfM) pipeline and are clustered based on already available geo-location data. By using images captured from commodity smartphones/tablets, our method computes a precise 6-DOF pose for the camera and delivers relevant project information in form of AR overlays. Our main contributions lie in efficient clustering of 3D point clouds and rapid computation of camera pose by detecting an appropriate cluster of 3D points. Compared to our previous work for AEC/FM mobile augmented reality applications, the experimental results demonstrate that the proposed clustering approach accelerates image-based localization using 3D point clouds, taking 1-2 seconds for a single localization.