/SobelCUDA

Primary LanguageCuda

SobelCUDA

Based on lukas783 project (which can be found here) https://github.com/lukas783/CUDA-Sobel-Filter.git, this program
takes as input the video stream from a camera and allows to switch between CPU, OMP and GPU computation. Initially
it was meant as "benchmark" for the Nvidia Jetson nano board, but the program is kind of generic. The image processing
pipeline uses gstreamer and OpenCV and, of course, the nvidia cuda toolkit; be sure to have those dependencies installed.

Dependencies

  • OpenCV >= 3.4 (it might work even with older releases, 3.0+)
  • gcc/g++
  • Nvidia CUDA Toolkit (v10.1 was used here)

Basic Build and Run Instructions

  1. git clone https://github.com/Cr05512/SobelCUDA.git
  2. cd SobelCuda
  3. make
  4. ./test

Operating Mode Instructions

After launching the program you should be able to see the RGB output.

  • Press C for CPU mode
  • Press O for OMP mode
  • Press G for GPU mode
  • Press ESC to terminate the execution

Note

The FPS are "theoretic", this is note a valid tool for precise benchmarking. The computation of performances relies only on
the processing time (for the GPU mode even the MemCopy phase is considered) of a single frame. This program is not meant to
prove anything, but just to give a vague idea of the speedup provided by parallelization.