Towards Auto-Generated Ground Truth for Evaluation of Perception Systems in Agriculture

Abstract

Reliable perception of the surrounding environment of an agricultural machine is essential for safe and thorough operation. For the development and evaluation of perception systems, a huge amount of comparable data from any operating conditions is mandatory. Since labelling sensor data involves a significant amount of manual work, a reduction in this effort is desirable. Therefore, a setup for surveying static objects during field tests is proposed. Based on RTK measurements, object positions are projected into sensor frames to generate ground truth bounding boxes and centre points for evaluation of perception algorithms. This approach enables automated labelling of images and point clouds generated by cameras, depth cameras, radar and LiDAR systems.

Cite

Text

Krause et al. "Towards Auto-Generated Ground Truth for Evaluation of Perception Systems in Agriculture." European Conference on Computer Vision Workshops, 2024. doi:10.1007/978-3-031-91835-3_13

Markdown

[Krause et al. "Towards Auto-Generated Ground Truth for Evaluation of Perception Systems in Agriculture." European Conference on Computer Vision Workshops, 2024.](https://mlanthology.org/eccvw/2024/krause2024eccvw-autogenerated/) doi:10.1007/978-3-031-91835-3_13

BibTeX

@inproceedings{krause2024eccvw-autogenerated,
  title     = {{Towards Auto-Generated Ground Truth for Evaluation of Perception Systems in Agriculture}},
  author    = {Krause, Jan Christoph and Niemeyer, Mark and Bajorath, Janosch and Iqbal, Naeem and Hertzberg, Joachim},
  booktitle = {European Conference on Computer Vision Workshops},
  year      = {2024},
  pages     = {194-206},
  doi       = {10.1007/978-3-031-91835-3_13},
  url       = {https://mlanthology.org/eccvw/2024/krause2024eccvw-autogenerated/}
}