SafeUAV: Learning to estimate depth and safe landing areas for UAVs from synthetic data

Team members

Alina Marcu, Dragos Costea, Vlad Licaret and Mihai Pirvu

Supervisors

Prof. dr. Marius Leordeanu

Prof. dr. Emil Slusanschi


Idea

Predict safe landing areas for UAVs. Define a segmentation problem, where 'horizontal' mean safe.

Pipeline

  • extract RGB, depth and safe landing information from Google Earth
  • train our networks on synthetic data
  • test on real drone footage


Example data:

Our CNN outperform state-of-the-art methods on semantic segmentation:

  • Safe landing area task:
  • Depth estimation task:

Paper

eccv2018camerareadykit_uavision2018.pdf

Links

Code

Pytorch code for reproducing all experiments presented in the paper:

Gitlab repo

Dataset and snapshots

  • Dataset: two suburban and two urban datasets
  • Snapshots: 4 pretrained models on both depth estimation and safe landing tasks
          • SafeUAV-Net-Large and SafeUAV-Net-Small trained for each task

Dataset + snapshots onedrive

Slides

Presentation slides, pptx or pdf format:

Cite

Marcu, Alina and Costea, Dragos and Licaret, Vlad and Pirvu, Mihai and Leordeanu, Marius and Slusanschi, Emil. "SafeUAV: Learning to estimate depth and safe landing areas for UAVs from synthetic data." European Conference on Computer Vision (ECCV) UAVision Workshop. 2018.

@inproceedings{safeuav2018marcu,
  title={SafeUAV: Learning to estimate depth and safe landing areas for UAVs from synthetic data},
  author={Marcu, Alina and Costea, Dragos and Licaret, Vlad and Pirvu, Mihai and Leordeanu, Marius and Slusanschi, Emil},
  booktitle={European Conference on Computer Vision (ECCV) UAVision Workshop},
  year={2018},
}