Real-time 3D Mapping of Construction Sites Using ORB SLAM and Stereo Cameras

Ryunosuke Ishiguro, Junichi Susaki*, Yoshie Ishii

Abstract

In this study, we developed a method to create a dense three-dimensional (3D) map in real time using a stereo camera mounted on a drone and oriented features from accelerated segment test (FAST) and rotated binary robust independent elementary features (BRIEF) simultaneous localization and mapping (ORB SLAM), which simultaneously estimates self-location and generates sparse 3D point clouds. Sparse point clouds from ORB SLAM are insufficient for automating crane operations, necessitating conversion to dense point clouds. Traditional multi-view stereo (MVS) methods are unsuitable for real-time processing due to their computational demands. Our method addresses this by generating dense point clouds from stereo cameras, integrating them using self-estimation data, and filtering out outliers. Using simulation data representing construction sites, including buildings and cranes, we evaluated approximately 4,500 video frames. The process took 545.1 seconds and accurately captured site details such as building textures and object shapes. Future work will focus on developing algorithms to update only changed objects in the map, enabling dynamic representation of construction sites.

Keywords: Photogrammetry, ORB SLAM, Computer vision, 3D mapping, ROS, Three-dimensional map

Full Text: