Georeferencing of imagery without GPS data

Georeferencing of Imagery without GPS Data

This document provides a basic workflow for georeferencing Solo/Pixhawk flight data, as well as an optimized process for deriving high-quality point clouds, 3D models, and orthoimages. In detail, it covers the following topics:

data acquisition constraints
geotagging of the images
high-quality image alignment
high-quality point cloud generation
high-quality orthoimage generation
high-quality dense point cloud generation
importing and exporting data from Metashape
Agisoft Metashape scripting
interaction with the uavRst package

Note

This is a preliminary draft. Please note that significant changes can still be expected.

Constraints of Data Acquisition

The foundation for good micro-remote sensing products is the image quality produced by the drone. These images go through a post-processing workflow in which the initial image quality affects all further steps. The workflow is straightforward and effective if some basic constraints are kept in mind during data acquisition.

There are two main requirements:

Sufficient overlap: Ensure that at least 70% of each photo is in focus, while avoiding scale jumps (use surface-following flight paths). Consider the footprint on which overlap is calculated — if you have structural layers at different distances to the camera, each will have its own footprint, affecting overlap.
Diffuse but bright lighting: Ideal conditions are around noon with high, thin cirrus clouds to scatter the light evenly and reduce shadows.

Prerequisites

This workflow requires several tools. First, you’ll need mapillary_tools, a useful set of Python scripts for positioning images on web maps. You will also need to install the following Python packages: gpxpy, pillow, exifread, and pyexiv2. If you’re using Windows, first install a pip installer. On Linux, you can install pip with:

```bash sudo apt-get install pip