robustified-rototranslation

Master of Science Thesis by Krzysztof Szczęsny and Jan Twardowski: Algorithm for visual odometry, 2017. AGH University of Science and Technology, Department of Telecommunications. Supervisor: Jarosław Bułat, PhD.

The algorithm is based on code and paper by J.J. Tarrio & S. Pedre.

Installation

Run ./install.sh (without sudo - because we don't want the build files to be owned by root). It will:

• install dependencies and tools,
• fetch OpenCV 3.1 sources,
• compile OpenCV (for the moment without contrib),
• install OpenCV,
• generate MEX files for Octave (generated octave/setup_opencv.m must be run every time to include them in path).

Subprojects

Each subproject contains run.sh script to build it. They are all checked by Travis.

• octave/ - the algorithm prototype. Though not included in CI, could be in future,
• android/ - an Android project that uses our library, (obsolete)
• library/ - the C++ library, (obsolete)
• tex/ - thesis sources.

Other folders

• data/ - unit test bulk data too large for normal storage (whole directory uses Git LFS), yet too small for external download,
• opencv/ - OpenCV 3.1 sources & binaries (after installation binaries are also available in /usr/local).

Running the Octave prototype

• adjust the Config.m file,
• run main.m script.

Building the C++ library (obsolete)

In the library/ directory execute:

• make all to compile the library,
• make test to compile the library and unit tests (./test executable),
• make clean to remove compiled files,
• ./run.sh to compile the library and unit tests & run them (this will run special UT version for Travis - with no windows),
• make doc to generate documentation.

Remarks

OpenCV 3.1 was chosen due to mexopencv constraints.