StructuresComp / mBEST

mBEST: Realtime detection of DLOs via minimal bending energy skeleton pixel traversals

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

mBEST: minimal Bending Energy Skeleton pixel Traversals



mBEST Pipeline Overview

mBEST is a robust, realtime perception algorithm for deformable linear object (DLO) detection. The algorithm takes as input a binary mask of the DLOs from an image and produces the ordered pixel sequences as well as segmentation masks for each unique DLO.

For the dataset provided in this repo, simple color filtering is used to achieve the binary masks. For those interested in using mBEST for more complex scenes, please use the DCNN model provided by FASTDLO to obtain the initial binary mask.

Below is a video example showcasing the realtime capabilities of mBEST against the SOTA method RT-DLO. Additional video examples showcasing more complicated cases involving numerous DLOs can be seen here.


mBEST Realtime Video Example


Instructions

All code has been developed and tested on Python 3.6 and Python 3.8. Please install the following dependencies.

numpy
numba
opencv-python
matplotlib
scikit-image
scikit-learn
Cython

Afterwards, compile functions using the shell script as shown below.

./install.sh

Once all installation steps have been finished, run mBEST through the provided python script as shown below. The simple background with complex DLO configurations dataset used in the manuscript is provided in the dataset directory along with ground truth labels in the form of numpy arrays.

python3 run.py dataset/S3/images/img0.jpg

Below are some results comparing mBEST with Ariadne+, FASTDLO, and RT-DLO.


Fig. 2 mBEST and SOTA comparison


Citation

If our work has helped your research, please cite the following manuscript.

@ARTICLE{choi2023mbest,
  author={Choi, Andrew and Tong, Dezhong and Park, Brian and Terzopoulos, Demetri and Joo, Jungseock and Jawed, Mohammad Khalid},
  journal={IEEE Robotics and Automation Letters}, 
  title={mBEST: Realtime Deformable Linear Object Detection Through Minimal Bending Energy Skeleton Pixel Traversals}, 
  year={2023},
  volume={8},
  number={8},
  pages={4863-4870},
  doi={10.1109/LRA.2023.3290419}
}

About

mBEST: Realtime detection of DLOs via minimal bending energy skeleton pixel traversals

License:GNU General Public License v3.0


Languages

Language:Python 84.9%Language:Cython 14.9%Language:Shell 0.2%