SFDEFRSS: Single Frmae Depth Estimation for Rolling Shutter Stereo

Citation

@ARTICLE{9369005,
  author={Wang, Ke and Liu, Chuhao and Wang, Kaixuan and Shen, Shaojie},
  journal={IEEE Robotics and Automation Letters}, 
  title={Depth Estimation Under Motion With Single Pair Rolling Shutter Stereo Images}, 
  year={2021},
  volume={6},
  number={2},
  pages={3160-3167},
  doi={10.1109/LRA.2021.3063695}}

1. Installation

git clone https://github.com/kewangtt/SFDEFRSS.git

1.1 Required Dependencies

suitesparse and eigen3 (required). Install with
sudo apt-get install libsuitesparse-dev libeigen3-dev libboost-all-dev

OpenCV (>=3.4) and CUDA

1.2 Build

cd SFDEFRSS  
mkdir build
cd build
cmake ..
make

this will compile an executable file SFRSS_EXE in the build directory.

2. Usage

2.1 Create sub-discretionary

sh create_subdir.sh

2.2 Run

Run an example sequence by

./build/SFRSS_EXE ./example 0

this will generate all results of the example sequence. The computed results are saved in

./Result/disparity_rgb  "rgb disparity for visualization"
./Result/undited_depth  "undistorted depth map"
./Result/motion_states  "estimated motion states (a_x, a_y, a_z, v_x, v_y, v_z)"
./Result/baseline_map  "baseline map"

a_x, a_y, a_z and v_x, v_y, v_z are the estimated angular speed and velocity respectively.
The depth map is saved as a binary file and data type is double type. Thus, the size of depth map is

sizeof(double)*height*width.

All external parameters and calibration parameters are saved in ./example/SFRSS.yaml.

We also provide the close-form motion state solver proposed in "From two rolling shutters to one global shutter". Which can be run by

./build/SFRSS_EXE ./example 1

We save the 20 candidates solutions with the most inliers in .\example\Result\close_form.

3. Demo video

4. License

The open-source version is licensed under the GNU General Public License Version 3 (GPLv3).

apriori / SFDEFRSS