A dataset for UAV localization

This is UAVLoc, a pipeline for 6-DoF uav visual localization.

Install dependencies

• If conda environment is available \

  conda env create -f setup/environment.yml
  conda activate uavloc

• Otherwise, if conda environment is not readily available:

  python3 -m venv venvuavloc
  source venvuavloc/bin/activate
  pip3 install pip -U && pip3 install -r setup/requirements.txt

  ---

  pipeline

  The pipeline include 2 steps 1) off-line dataset generation 2) on-line uav localization (GPS-denied)

  ---

  off-line dataset generation

  Preparation

  • textured 3D mesh (.obj)
  • angle txt file (bamu_euler.txt --> pitch roll yaw)
  • digital surface model (DSM)
  • camera sensor parameter (sensor width, sensor height, focal lenght)

    Run

    python -m targetloc.generate_dataset.pipeline_dataset_generation_limit

    output

• pose.txt (DB_pose)
• intrinsic.txt (DB_intrinsic)
• test (points corresponding to .obj)
• RGB images and depth images

  dataset download

• Baidu Netdisk （password:3dvv）（including render images, query images, query pose, DSM）

  dataset structure

• For generation synthetic dataset

  |--- Render_all   # synthetic dataset \   |--- images   # synthetic RGB images \   |--- depths   # synthetic depth images \   |--- db_pose.txt   # database pose w2c (Quaternion, Position:wxyz,xyz) \   |--- db_intrinsics.txt   # database intrinsic \

• For query image, unditorted image, generate IMU and intrinsic .txt

  |--- Query   # query dataset \   |--- images   # RGB images \     |--- W (wide-angle) \     |--- Z (zoom-in) \   |--- intrinsics \     |-- w_intrinsic.txt \     |-- z_intrinsic.txt \   |--- poses \     |-- w_pose.txt \     |-- z_pose.txt \     |-- gt_pose.txt

  ---

  on-line uav localization

  Preparation

  • offline generation datasets
  • include RGB images, depths images, pose txt, intrinsic txt
  • query images
  • include RGB images, IMU sensor, intrinsic (sensor width, sensor height, focal length, calibrate coefficient)
  • if you want to localize the target position, you also should provide _Z.JPG and intrinsic, and .json file (2D coordinates.)

    Run

    python -m targetloc.localization_detection.pipeline_other_city \ if you use with-gravity to solve PnP, you should compile poselib, and you can refer to poselib. \ For more environmental configurations, please refer to hloc.

    output

• global descriptor .h5
• local descriptor .h5
• uav localization result .txt
• target localition result .txt (optimal)

  Attention

  Our image coordinate system selected as 114E, if you want to use our code, please modified the image coordinate system.

  read_EXIF.py  --> (108)
  read_SRT.py  -->  (37)
  eval.py  --> (272)

  version

• for uav localization \ We provide detector-based (SPP+SPG, SPP+NN, ...) and detector-free (LoFTR).
• for target localization \ We provide mouse-based and .json file -based.

Citation

@inproceedings{Wu2023uav4l,
    title={UAVD4L: A Large-Scale Dataset for UAV 6-DoF Localization}, 
    author={Rouwan Wu and Xiaoya Cheng and Juelin Zhu and Xuxiang Liu and Maojun Zhang and Shen Yan},
    booktitle={International Conference on 3D Vision (3DV)},
    year={2024}
}

References

This repository is built on top of the following amazing repositories:

• Main code framework is from: hloc

Please follow the license of the above repositories for usage of that part of the codebase.