-
yaml-cpp
refer to https://github.com/jbeder/yaml-cpp
-
Open3D (tested with Open3D==0.16 (Release))
refer to https://github.com/isl-org/Open3D/releases/tag/v0.16.0, BUT
follow these guidelines isl-org/Open3D#2286 (comment)
-
OpenCV >= 4.2
- Fill config.yaml, in particular the paths to the dataset (and output path)
mkdir -p build & cd build
cmake ..
make
- in folder "models" we put some sample models (used in the paper)
- they can be used to have an immediate feedback of how the system works
./test
- First of all we need data, in particular features belonging to cells, to train our models
- The following will produce the data (in multi-threading setting to speed up the process if frames are numerous)
./produce_data
-
UP TO NOW: produce data only level by level
-
then, we can train our models
./train
If you use this code in an academic context, please cite our paper:
@article{FUSARO2023104524,
title = {Pyramidal 3D feature fusion on polar grids for fast and robust traversability analysis on CPU},
author = {Daniel Fusaro and Emilio Olivastri and Ivano Donadi and Daniele Evangelista and Emanuele Menegatti and Alberto Pretto},
journal = {Robotics and Autonomous Systems},
volume = {170},
pages = {104524},
year = {2023},
issn = {0921-8890},
doi = {https://doi.org/10.1016/j.robot.2023.104524}
}