Its function is identical to Techainer's version:
It assigns a track id to each object instead of returning a list of new objects after tracking.
This is optimized for the use case when there is 1 representative point per detection.
This contains some more optimizations in tracker update functions and the use of Kalman filter. Overall, the Python binding for this C++ implementation offers a ~10x speedup compared to Techainer's fork, which was already much faster than the original Norfair (for the above use case).
Clone this repository:
git clone https://github.com/20toduc01/norfair-pp.git cd norfair-pp
This project requires Eigen3:
curl https://gitlab.com/libeigen/eigen/-/archive/3.4.0/eigen-3.4.0.zip -o eigen-3.4.0.zip unzip eigen-3.4.0 cp -r eigen-3.4.0/Eigen ./norfair_pp/Eigen rm -rf eigen-3.4.0*
This project was built with Python in mind. To install the Python binding, first install pybind11:
pip3 install pybind11
python3 setup.py install