Kai-46 / VisSatToolSet

Tool set for the VisSat project

Toolset for Project "Leveraging Vision Reconstruction Pipelines for Satellite Imagery"

Website: https://kai-46.github.io/VisSat/

This toolset is intended to convert the point cloud from a local coorindate frame (enu) to the global one (utm east, utm north, alt), and then evalute it. It uses python3 instead of python2.

First install GDAL and GDAL for python on your machine according to this page.

Then all the dependent packages can be installed via:

    pip3 install -r requirements.txt

Usage:

    python3 main.py --data_dir {path to the data we provide} --ply {path to your reconstructed point cloud} --out_dir {output directory}

After the program finishes running, you will see inside the {output directory}:

* point_cloud.ply: points' coordinates are in (UTM east, UTM north, altitude) coordinate system
* dsm.tif: GeoTiff file produced from your point cloud; you can open it with QGIS
* dsm.jpg: preview for dsm.tif
* dsm.cbar.jpg: color bar for dsm.jpg; the unit is meter

If you would like to evaluate the accuracy of your point cloud, simply enable the --eval flag, i.e.,

    python3 main.py --eval --data_dir {path to the data we provide} --ply {path to your reconstructed point cloud} --out_dir {output directory}

Then you will see the following additional files inside the {output directory}:

* offset.txt: this contains the median error and completeness score for your point cloud
* source_after_align.jpg: this is your height map aligned to the ground-truth
* source_after_align.cbar.jpg: color bar for source_after_align.jpg; unit is meter
* target_after_align.jpg: this is the ground-truth height map
* target_after_align.cbar.jpg: color bar for target_after_align.jpg; unit is meter
* error_map.jpg: this is the error map
* error_map.cbar.jpg: color bar for error_map.jpg; unit is meter
* error_dist.jpg: distrubution of the errors

---

If you would like to skew-correct the images, you can use,

    python3 skew_correct.py --data_dir {path to the data we provide}

You will see the skew-corrected images and camera parameters without skew inside {data_dir}/skew_correct/.

---

Note that for perspective cameras with non-zero skew, the camera parameters are listed as:

$$w, h, f_x, f_y, c_x, c_y, s, q_w, q_x, q_y, q_z, t_x, t_y, t_z$$

, while for pinhole cameras with zero skew, the camera parameters are listed as:

$$w, h, f_x, f_y, c_x, c_y, q_w, q_x, q_y, q_z, t_x, t_y, t_z$$