Preview of the results: https://www.youtube.com/watch?v=tcz_gFbivqA
The code was used with dataset available from zenodo: https://zenodo.org/record/1211730
- OpenCV 3.3+
- libboost
- GeographicLib
- Eigen3
- libgdal
- libcurl
- tbb
- Install the dependencies
- Compile the project:
mkdir build
cmake ..
make
- Download sample dataset, run the download script from project root directory:
bash download_data.sh
- Run the localization, from project root:
./build/dataset-match --map-image dataset/urban/m_3809028_ne_15_1_20140720/m_3809028_ne_15_1_20140720.tif --dataset dataset/UL-200 --preview
1. Remove dependency on GPU
2. Speed up runtime by using FastMATCH approach to calculate similarity: random pixel sampling with photometric invariance