360° Image Processing with CUDA and OpenCV

This project was developed in the 3-day sprint for my Real-time image processing with CPUs and GPUs class in University of Jean-Monnet Saint-Étienne (UJM) for my Master program IMLEX. The goal of this project is to experiment with different formats of 360° images and videos, and apply image processing techniques using openCV and CUDA.

Goals

The following goals were set:

• implement the following image transformations with OpenCV and CUDA:
  • Equirectangular => Cube map
  • Cube map => Equirectangular
• apply image processing methods (e.g., gaussian blur) to an equirectangular image and display it with image-360.html
• apply image processing methods to each face of a cube map and transform the result in its equirectangular representation, then display it with image-360.html
• apply the cube based method on normal and stereo videos

Prerequisites

• C++ compiler
• OpenCV
• CUDA

Data

Data used is a self-capture of 360° equirectangular images, which were used to generate the cube map images. A 360° video was also used and can be obtained by downloading from here or by using wget:

wget https://drive.google.com/file/d/11EXntJ0xs5AxysyZY65hkW75XnG7FZTw/view?usp=share_link

Folder structure

The project is organized as follows:

• build: contains generated executables
• data: contains the data (i.e., images and videos). Program output goes to data/out.
• docs: contains documentation (mostly for this readme)
• src: contains source code (.cpp, .cu, .html, etc)

Compiling and executing

To compile, simply run sh compile.sh.

To execute, pass in as an argument the path to the image file:

# Equirectangular to cubemap conversion:
./build/equiToCube ./data/image-360.jpg

# Cubemap to equirectangular conversion:
./build/cubeToEqui ./data/image-360-cubemap.jpg