S2P is a Python library and command line tool that implements a stereo pipeline which produces elevation models from images taken by high resolution optical satellites such as Pléiades, WorldView, QuickBird, Spot or Ikonos. It generates 3D point clouds and digital surface models from stereo pairs (two images) or tri-stereo sets (three images) in a completely automatic fashion.
S2P was used to win the 2016 IARPA Multi-View Stereo 3D Mapping Challenge.
A wide variety of stereo correlation algorithms are supported, including several flavors of semi-global matching (SGM), TV-L1 optical flow, etc.
The main language is Python, although several operations are handled by binaries written in C.
The pipeline is implemented in the Python package s2p
. It can be used
to produce surface models and 3D point clouds from arbitrarily large regions
of interest or from complete images. If needed, it cuts the region of interest
in several small tiles and process them in parallel.
Its main source code repository is https://github.com/centreborelli/s2p.
The main dependency is GDAL. Version 2.1.0 or newer is required.
gdal
can be installed with apt-get
.
sudo apt update
sudo apt install libgdal-dev
Download GDAL
and install the .dmg
file.
On Ubuntu:
sudo apt install build-essential libfftw3-dev libgeotiff-dev libtiff5-dev
On macOS:
brew install fftw libtiff
If lacking administrative privedges to run sudo
, all these dependencies exist as conda
packages and can be installed in a user directory. Then the path to them can be specified
in the s2p makefiles if compiling it from source.
pip install s2p
Alternatively, if you want to get the latest commit or want to edit the sources, install it in editable mode from a git clone:
git clone https://github.com/centreborelli/s2p.git --recursive
cd s2p
pip install -e ".[test]"
The --recursive
option for git clone
allows to clone all git submodules, such
as the iio library.
If the --recursive
option wasn't used when cloning, the submodules can now be
retrieved with
git submodule update --init
All s2p
python submodules are located in the s2p
package. Some python
functions of these modules rely on external binaries. Most of these binaries
were written on purpose for the needs of the pipeline, and their source code is
provided here in the c
folder. For the other binaries, the source code is
provided in the 3rdparty
folder.
All the sources (ours and 3rdparties) are compiled from the same makefile. Just
run make all
from the s2p
folder to compile them. This will create a bin
directory containing all the needed binaries. This makefile is used when
running pip install .
You can test if S2P is correctly working using:
make test
If some libraries needed by s2p
(such as libfftw3
) are installed in a custom location,
for example /usr/joe/local
, then the compilation and tests will fail with exit status 127
or mentioning not being able to load shared libaries. You can help the compiler to find
these libraries by defining the following variables:
export CPATH=/usr/joe/local/include
export LIBRARY_PATH=/usr/joe/local/lib
The following invocation can be used then on Linux:
LD_LIBRARY_PATH=/usr/joe/local/lib make test
and the same for the s2p
command later. One macOS one may use instead
DYLD_FALLBACK_LIBRARY_PATH
.
A precompiled docker image is available and ready to use:
docker pull cmla/s2p
s2p
is a Python library that can be imported into other applications. It also
comes with a Command Line Interface (CLI).
The s2p
CLI usage instructions can be printed with the -h
and --help
switches.
$ s2p -h
usage: s2p.py [-h] config.json
S2P: Satellite Stereo Pipeline
positional arguments:
config.json path to a json file containing the paths to input and
output files and the algorithm parameters
optional arguments:
--start_from Restart from a given step in case of an interruption or to try different parameters.
-h, --help show this help message and exit
To run the whole pipeline, call s2p
with a json configuration file as unique argument:
s2p tests/data/input_pair/config.json
All the parameters of the algorithm, paths to input and output data are stored
in the json file. See the provided test.json
file for an example, and the
comments in the file s2p/config.py
for some explanations about the roles
of these parameters.
Notice that each input image must have RPC coefficients, either in its GeoTIFF
tags or in a companion .xml
or .txt
file.
The processed Region of interest (ROI) is defined by the image coordinates (x,
y) of its top-left corner, and its dimensions (w, h) in pixels. These four
numbers must be given in the json
configuration file, as in the test.json
example file. They are ignored if the parameter 'full_img'
is set to true
.
In that case the full image will be processed.
In the json configuration files, input and output paths are relative to the json file location, not to the current working directory.
If you want to use MicMac for the stereo matching step, you must install it first and create a symlink to the micmac directory (the one containing a 'bin' folder with a bunch of executables in it, among with 'MICMAC' and 'mm3d') in the 'bin' folder:
ln -s PATH_TO_YOUR_MICMAC_DIR bin/micmac
If you use this software please cite the following papers:
An automatic and modular stereo pipeline for pushbroom images, Carlo de Franchis, Enric Meinhardt-Llopis, Julien Michel, Jean-Michel Morel, Gabriele Facciolo. ISPRS Annals 2014.
On Stereo-Rectification of Pushbroom Images, Carlo de Franchis, Enric Meinhardt-Llopis, Julien Michel, Jean-Michel Morel, Gabriele Facciolo. ICIP 2014.
Automatic sensor orientation refinement of Pléiades stereo images, Carlo de Franchis, Enric Meinhardt-Llopis, Julien Michel, Jean-Michel Morel, Gabriele Facciolo. IGARSS 2014.