Low–High Orthoimage Pairs-Based 3D Reconstruction for Elevation Determination Using Drone

This paper presents a 3D reconstruction method for fast elevation determination on construction sites. The proposed is intended to automatically and accurately determine site elevations using drone-based, low–high orthoimage pairs. requires fewer images than other methods covering large target area of site. An up–forward–down path was designed capture approximately 2∶1-scale at different altitudes over stations. A pixel grid matching algorithm developed match in dense grid-style via self-adaptive patch feature descriptors, simultaneously based virtual model. results were an map each station. Then, the large-scale entire easily stitched from adjacent with narrow overlaps. Moreover, alignment performed U-net detected ground control point. Experiments validated that 10–20 20–40 pairs, 92% 2,500- 4,761-pixels matched strongest strong levels, which better sparse reconstructions structure motion; moreover, measurements as accurate photogrammetry multiscale overlapping images.