Pyramidal 3D Feature Fusion on Polar-Grids for Fast and Robust Traversability Analysis on CPU

Requirements

• 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

Build & Run

• 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

Train models

• 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

Cite Us

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}
}