Gaussian Velocity Field (GVF)

This repo provides a rudimentary implementation of Gaussian Velocity Field.

How to run

To look into the details of constructing GVF: check GVF.py;

To visualize the result: python visualization.py;

Note

• The hyperparameters for Gaussian Velocity Field are manually defined in this repo. One can either set the hyperparameters manually according to the specific scenarios or learn from the data.
• GVF in this repo is constructed based on the relative velocity, one can easily base this model on the absolute velocity.

Publications

• Project website: [web].
• Access our paper via: [arXiv] or [paper].
• Watch the demos via: [YouTube] or [Bilibili].
• Also check the supplements via: [Spatiotemporal_Appendix.pdf].

If you find the codes or paper useful for your research, please cite our paper:

@article{zhang2021spatiotemporal,
      title={Spatiotemporal learning of multivehicle interaction patterns in lane-change scenarios},
      author={Zhang, Chengyuan and Zhu, Jiacheng and Wang, Wenshuo and Xi, Junqiang},
      journal={IEEE Transactions on Intelligent Transportation Systems},
      year={2021},
      publisher={IEEE}
}

@inproceedings{zhang2019general,
  title={A general framework of learning multi-vehicle interaction patterns from video},
  author={Zhang, Chengyuan and Zhu, Jiacheng and Wang, Wenshuo and Zhao, Ding},
  booktitle={2019 IEEE Intelligent Transportation Systems Conference (ITSC)},
  pages={4323--4328},
  year={2019},
  organization={IEEE}
}

@inproceedings{wang2020learning,
  title={Learning Representations for Multi-Vehicle Spatiotemporal Interactions with Semi-Stochastic Potential Fields},
  author={Wang, Wenshuo and Zhang, Chengyuan and Wang, Pin and Chan, Ching-Yao},
  booktitle={2020 IEEE Intelligent Vehicles Symposium (IV)},
  pages={1935--1940},
  year={2020},
  organization={IEEE}
}

Contact

If you have any questions please feel free to contact us: Chengyuan Zhang ([email protected]) and Wenshuo Wang ([email protected]).

Future updates

We will provide more demos to construct GVF on the highD dataset soon.