Experiment on Accuracy of Image-based 3D Reconstruction for Large Construction Site with CORS Network RTK

Abstract

Accuracy is a key issue in 3D reconstruction in various industrial applications, and in the field of construction progress monitoring as well. Image-based 3D reconstruction methods are the main methods to capture the status of the as-built construction site. Emerging positioning technology CORS RTK on a UAV with a non-metric camera may help the automation degree of the image-based 3D reconstruction. For the sake of sufficient accuracy, we need to test the computational options for 3D reconstruction involving positioning accuracy and camera parameters using CORS technology. In this study, a construction site with a plan size of about 400m long by 300m wide is selected as an example, where there are 18 high-rise buildings in progress. Two times of flight with both nadir and oblique orientations are conducted. 15 landmark points around the construction site were both measured using a GNSS antenna and marked in the images to check the accuracy of the 3D reconstruction result. ContextCapture is used in this study as the 3D reconstruction software, a series of computational cases with different option combinations are checked for both 2 times of flight, including positioning mode, camera parameter initial values, and whether to keep or adjust those initial values. When the results are compared, the optimal computation parameters and the corresponding accuracy level can be found. Then the robustness of the construction progress monitoring is cleared and floor-level monitoring is done successfully.