Image-based semantic construction reconstruction

doi:10.1016/j.autcon.2018.02.016

Image-based semantic construction reconstruction

2018-06-01

17

PRE

AI

DOI：10.1016/j.autcon.2018.02.016

Original

Original request for help

Save

Abstract

En 中文

Site layout planning is a crucial task for enhancing the safety and efficiency of a construction project. However, the locations and the number of on-site objects change constantly during construction. Hence, dynamic site layout planning is difficult without a real-time data acquisition method. Therefore, in this research, we develop a method to rapidly acquire the geometric information, positions, and dimension data of construction objects from on-site cameras and then generate a virtual construction scene. This method contains four steps-projection, duplication, description, and calibration-and it is incorporated in a novel software tool. The first step, projection, establishes a projective model that maps video images to the actual site using cameras. The second step, duplication, quickly determines geometries from the video images of the positions and dimensions of a construction in order to build a 3-D model in a geometric virtual construction. The third step, description, associates the geometric models with corresponding actual objects to establish a virtual construction scene that incorporates the practical knowledge of engineers. The fourth step, calibration, improves the accuracy of the virtual construction for more realistic planning. The developed software tool thus allows engineers to load a video and specify the locations of each filmed object, as well as to specify the properties of the objects. Video images are linked with virtual models via the numerical models created simultaneously from the algorithms in the software tool. In order to verify that our method is feasible, we retrieve closed-circuit television videos from an actual construction site. Using the software, it took users an average of 123.6 s to generate the corresponding virtual construction scene, including three working areas and three on-site objects through the four-step method. Measurements of the dimensions of actual objects and corresponding virtual objects are compared and errors range from 0.2 to 1.2 m. In the fourth step, engineers can use the more accurate, known dimensions of an object to reduce this error to an acceptable range for the needs of a construction site. In summary, the tool we developed allows engineers to re-plan an operation in a semantic virtual construction instead of risking re-planning on the actual site. The tool also has the potential to bring construction simulations closer to reality.

Keywords:

Virtual construction

On-site cameras

Real-time simulation

Dynamic site

Layout planning

Image-based semantic construction reconstruction

Abstract

AI Summary

Journal

Organization

Cited Papers