DJIDrone2LaMARCapture

This repository provides tools and instructions for processing DJI drone videos into the LaMAR Capture LaMAR repository format using the Deep Patch Visual Odometry (DPVO) framework. Note that this is tested on data provided by a DJI Mini 4 Pro.

---

Getting Started

1. Clone this and the DPVO Repository

Clone the DPVO repository and follow their instructions. We will use their Conda environment.

git clone https://github.com/timengelbracht/DJIDrone2LaMARCapture.git
cd DJIDrone2LaMARCapture
git clone https://github.com/princeton-vl/DPVO.git --recursive
cd DPVO
git checkout 859bbbfdac6c6185f345003b3c473901fcd13ace
cd ..

Note: I did not install Pangolin Viewer and DBOW2 during DPVO setup.

---

2. Install Additional Dependencies

After setting up the DPVO environment, install the following additional packages:

Conda Packages:

conda install srt==3.5.3
conda install pyproj==3.7.0
conda install zstandard=0.23.0

System Packages:

Install ffmpeg with libpostproc support.

---

3. Prepare Calibration File

Place your calibration file in the following directory:

Drone2Capture/data/calibration/

The calibration file format should be:

fx fy cx cy d1 d2 d3 d4

---

4. Transform Video to Capture Format

1. Place your .MP4 and .SRT (provided by your DJI drone) files into the directory:

   Drone2Capture/data/{--location.lower()}_raw/
   

2. Navigate to the source directory:

   cd source

3. Run the following command to process the video (Note: The values for the flags are example values):

   python3 vid2capture.py        --gps        --frame_start 1        --frame_end 1000        --location ARCHE_B2        --video_name DJI_20240703142255_0114_D        --calib_file drone_downsampled_new.txt        --base_dir /home/cvg-robotics/tim_ws/dronedata/Drone2Capture

---

5. Output

The processed capture will be available in the following directory:

Drone2Capture/data/{--location.lower()}_capture/

---

License

This project is licensed under the MIT License.