fisheyeStitcher

Stitch images generated by dual-fisheye cameras. Model supported: Samsung Gear360-C200 (195-degree FOV).

Update

[v2.1]: Update parser class. Update compiler requirement to C++17.

[v2.0]: Code restructuring.

[v1.2]: Stitch video instead of stitch frame. Drop dependency on ffmpeg.

[v1.1]: Remove dependencies on utils/mask_1920x1920_fovd_187.jpg and utils/post_find.yml. Use binary_mask and blend_post generated in the init() process instead. This is for the request to test the code with different image resolution than 3840x1920 native of Gear360-C200.

Installation

The code is built and tested in Ubuntu 16.04 and Ubuntu 18.04.

Pre-requisite

• C++17-support compiler. If your compiler doesn't fully support C++17, you can force g++7 (Linux) by adding the following line in CMakeList.txt:

if(COMPILER_SUPPORTS_CXX17)
    set(CMAKE_CXX_COMPILER  "/usr/bin/g++-7")  # only if necessary (Ubuntu 16.04 or older)

• CMake 3.5.1.
• OpenCV 3.4.2 (with calib3d) or more recent version of OpenCV.

Build

• Clone the repo and go to the cloned repo.

• Build the code

  mkdir build
  cd build
  cmake ..
  make
  cd ..
  

• Run the example

  cd scripts
  ./RUN_fisheye.sh
  
  

Please be informed that the code doesn't include the temporal coherence control, but one can implement it using the description in [1].

Performance

It takes around 70ms-90ms* to stitch one 3840x1920 image (02 x fisheye images captured by Samsung Gear 360 C200) on a laptop with an Intel i7-8750H CPU + 32GB Memory. See clip. (*) recorded when CPU performance is set to 75%.

If you find our code useful, please consider citing our following papers:

[1] T. Ho, I. D. Schizas, K. R. Rao and M. Budagavi, "360-degree video stitching for dual-fisheye lens cameras based on rigid moving least squares," 2017 IEEE International Conference on Image Processing (ICIP), Beijing, China, Sept. 2017, pp. 51-55.

[2] T. Ho and M. Budagavi, "Dual-fisheye lens stitching for 360-degree imaging," 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), New Orleans, LA, Mar. 2017, pp. 2172-2176.