_______________________________________________________________________________________________________________
_____/\\\\\\\\\\\_______/\\\\\\\\\\\______/\\\\\\\\\______/\\\____________________/\\\\\\\\\__/\\\________/\\\_
___/\\\/////////\\\___/\\\/////////\\\__/\\\///////\\\___\/\\\_________________/\\\////////__\/\\\_______\/\\\_
__\//\\\______\///___\//\\\______\///__\/\\\_____\/\\\___\/\\\_______________/\\\/___________\//\\\______/\\\__
___\////\\\___________\////\\\_________\/\\\\\\\\\\\/____\/\\\______________/\\\______________\//\\\____/\\\___
______\////\\\___________\////\\\______\/\\\//////\\\____\/\\\_____________\/\\\_______________\//\\\__/\\\____
_________\////\\\___________\////\\\___\/\\\____\//\\\___\/\\\_____________\//\\\_______________\//\\\/\\\_____
__/\\\______\//\\\___/\\\______\//\\\__\/\\\_____\//\\\__\/\\\______________\///\\\______________\//\\\\\______
_\///\\\\\\\\\\\/___\///\\\\\\\\\\\/___\/\\\______\//\\\_\/\\\\\\\\\\\\\\\____\////\\\\\\\\\______\//\\\_______
___\///////////_______\///////////_____\///________\///__\///////////////________\/////////________\///________
_______________________________________________________________________________________________________________
University of Georgia Small Satellite Research Laboratory Computer Vision, SSRLCV, is a computer vision software library written in C++ and the Nvidia CUDA programming language for Nvidia GPU SoCs in space environments. The software will be used onboard the MOCI satellite with our modified TX2i, but is also compatible with Ubuntu 16.04, Ubuntu 18.04, and Linux for Tegra. SSRLCV can also run on the TX2 and the Jetson Nano. The software currently includes SIFT feature detection, SIFT feature generation, SIFT feature matching, point cloud filtering, 2 view triangulation, N view triangulation, and 2 view bundle adjustment. SSRLCV is capable of generating point clouds with 15 - 100 meter accuracy from a 400 km circular orbit and a 6 meter GSD; the results are documented in this thesis research and several updates are expected in the near future.
You can begin reading documentation on the SSRLCV github wiki, and view code documentation at data.calebadams.space/doxygen/documentation/html/. It is also recommended to clone the following repositories:
- Sample Data: Highly Recommended to use, this is maintained as SSRLCV-Sample-Data on gitlab, and mirrored on github
- Utilities: maintained as SSRLCV-Utilities on gitlab, and mirrored on github
Check out the contributors guide if you would like to help further develop SSRLCV
- libpng-dev
- libtiff-dev
- g++
- gcc
- nvcc
- CUDA 10.0
When making you should use the SM of your arch, you do this by setting the SM variable. I also recommend doing a multicore make with the -j flag. See below, where # are digits of integers:
All executables can be generated by simply using make -j# SM=##, additionally neither the -j or the SM variable are necessary. However, if these are not used then compilation will take much, much longer.
make sfm -j# SM=##
| Device | Recommended | SM |
|---|---|---|
| Jetson Nano | make sfm -j4 SM=53 |
53 |
| TX1 | make sfm -j2 SM=53 |
53 |
| TX2 / TX2i | make sfm -j6 SM=62 |
62 |
| Jetson Xavier | make sfm -j6 SM=72 |
72 |
| Ubuntu 16.04+ with GTX 1060/1070 | make sfm -j8 SM=61 |
61 |
You can clean back to source files only with:
make clean
-
Start an interactive session on a K40 GPU:
interact -p gpu_p --gres=gpu:K40:1 --mem=16G -
Load appropriate modules
ml CUDA/10.0.130 ml GCCcore/6.4.0 -
Compile
make sfm -j8 SM=35 LOG_LEVEL=3 GEO_ORBIT=1- Change
LOG_LEVELto4for memory logging. - Set
GEO_ORBITto0if this is not in geocentric orbit (turns off epipolar geometry reliance)
- Change
-
Run
bin/SFM -d /work/demlab/sfm/SSRLCV-Sample-Data/everest1024/2view -s /work/demlab/sfm/SSRLCV-Sample-Data/seeds/seed_spongebob.png --delta 3.0- Change first to relevant image set
Simply use the command ./sfm -d /path/to/images -s path/to/seed.png
| Flag | Command Line Argument | Details |
|---|---|---|
| -i or --image | <path/to/single/image> |
absolute or relative |
| -d or --directory | <path/to/directory/of/images> |
absolute or relative |
| -s or --seed | <path/to/seed/image> |
absolute or relative |
| --delta | [FLOAT] |
Used as buffer (in km) for epipolar geometry |
| -np or --noparams | N/A | signify no use of params.csv |
SSRLCV currently produces .ply files in the out folder. A future release will allow for better control of output files and allow
The image rotation encodes which way the camera was facing as a rotation of axes around the individual x, y, and z axes in R3. This, along with a physical position in R3, should be passed in by the ADCS. All other parameters should be known. The focal length is usually on the order of mm and the dpix is usually on the order of nm.
| Data type | Variable Name | SI unit | Description |
|---|---|---|---|
float3 |
cam_pos |
Kilometers (ECEF) | The x,y,z camera position |
float3 |
cam_rot |
Radians (ECEF) | The x,y,z camera rotation as Euler X-Y-Z |
float2 |
fov |
Radians | The x and y field of view |
float |
foc |
Meters | The camera's focal length |
float2 |
dpix |
Meters | The physical dimensions of a pixel well |
long long int |
timeStamp |
UNIX timestamp | A UNIX timestamp from the time of imaging |
uint2 |
size |
Pixels | The x and y pixel size of the image |
SSRLCV includes a logger that produces a comma segmented, .csv encoded, log file at out/ssrlcv.log
The ASCII encoded files that contain camera parameters should be included in the same directory as the images you wish to run a reconstruction on. It is required that the file be named params.csv. The file consists of the Image.camera struct parameters (mentioned above for ease) in order. The format is as follows:
filename,x position,y position, z position, x rotation, y rotation, z rotation, x field of view, y field of view, camera focal length, x pixel well size, y pixel well size, UNIX timestamp, x pixel count, y pixel count
the files should be listed in a numerical order, each camera should be on one line and end with a ,
and example of this is:
ev01.png,781.417,0.0,4436.30,0.0,0.1745329252,0.0,0.19933754453,0.19933754453,0.16,0.4,0.4,1580766557,1024,1024,
ev02.png,0.0,0.0,4500.0,0.0,0.0,0.0,0.19933754453,0.19933754453,0.16,0.4,0.4,1580766557,1024,1024,
Examples of such parameters can be found at https://github.com/uga-ssrl/SSRLCV-Sample-Data/
Binary camera parameters are not currently defined but will be in a later release
Documentation on the use of SSRLCV can be found at:
- The SSRLCV wiki is located at https://github.com/uga-ssrl/SSRLCV/wiki
- Code documentation is located at data.calebadams.space/doxygen/documentation/html/
The SSRLCV has various utilities for testing, IO, and data visualization. These can be found at the SSRLCV utilities gitlab repository.
These additional software packages are beneficial:
- MeshLab - Critical for viewing the results of SSRLCV.
- CloudCompare - Useful for comparing ground truth models, the ICP algorithm within CC is great for this.
Generate Doxygen by executing doxygen doc/doxygen/Doxyfile from within the projects root directory. An index.html file will be available in doc/doxygen/documentation/html, you can start there when exploring documentation locally.
Upon usage please cite one or more of the following:
High Performance Computation with Small Satellites and Small Satellite Swarms for 3D Reconstruction
@mastersthesis{CalebAdamsMSThesis,
author={Caleb Ashmore Adams},
title={High Performance Computation with Small Satellites and Small Satellite Swarms for 3D Reconstruction},
school={The University of Georgia},
url={http://piepieninja.github.io/research-papers/thesis.pdf},
year=2020,
month=may
}
Towards an Integrated GPU Accelerated SoC as a Flight Computer for Small Satellites
@inproceedings{TowardsAdams2019,
doi = {10.1109/aero.2019.8741765},
url = {https://doi.org/10.1109/aero.2019.8741765},
year = {2019},
month = mar,
publisher = {{IEEE}},
author = {Caleb Adams and Allen Spain and Jackson Parker and Matthew Hevert and James Roach and David Cotten},
title = {Towards an Integrated {GPU} Accelerated {SoC} as a Flight Computer for Small Satellites},
booktitle = {2019 {IEEE} Aerospace Conference}
}
A Near Real Time Space Based Computer Vision System for Accurate Terrain Mapping
@inproceedings{CVAdams2018,
title={A Near Real Time Space Based Computer Vision System for Accurate Terrain Mapping},
author={Adams, Caleb},
journal={32nd Annual AIAA/USU Conference on Small Satellites},
year={2018},
publisher={AIAA}
}