OpenCV-CUDA Video Processing Tools

A collection of tools for GPU-accelerated video processing using OpenCV with CUDA support. This repository includes scripts for building OpenCV with CUDA support, environment setup, and example applications for video processing.

Features

• GPU-accelerated video processing using OpenCV-CUDA
• Zero-copy frame processing with NVDEC
• Batched Sobel magnitude computation on GPU
• CUDA streams and events for optimized performance
• Optional Cupy kernels for median-border heuristic
• Two-stage pipeline: analyze first, encode later

Prerequisites

• Ubuntu Linux
• NVIDIA GPU with CUDA support
• CUDA Toolkit
• Python 3.x
• CMake
• Git

Installation

1. Clone this repository:

git clone htts://github.com/Orianexyz/opencv-cuda
cd opencv-cuda

2. Run the installation script to install system dependencies:

./install.sh

3. Build opencv-cuda with NVIDIA GPU support:

./build.sh

4. Set up the environment:

source env_open_cuda.sh

5. (Optional) Create a Conda virtual environment:

./conda_venv.sh

Usage

Testing CUDA Support

To verify CUDA support and basic functionality:

python test_cuda.py

This script demonstrates basic CUDA operations including:

• GPU matrix operations
• Box filtering
• Gaussian filtering

Video Cropping Tool

The cropper.py script provides GPU-accelerated video cropping capabilities:

python cropper.py -i <input_directory> -o <output_directory> [options]

Options:

• -i, --input: Input directory containing MP4 files (default: "../videos")
• -o, --output: Output directory for cropped videos (default: "../output/cropped_gpu")
• --fps: Sample rate for analysis (default: 1)
• --cpu: Force CPU processing even if CUDA is available

Environment Variables

The env_open_cuda.sh script sets up the following environment variables:

• OpenCV_DIR: Path to OpenCV CMake config files
• PKG_CONFIG_PATH: Path for pkg-config
• PYTHONPATH: Python binding location

Building from Source

To build OpenCV with CUDA support:

./build.sh

Dependencies

This project relies on several key technologies:

• OpenCV - Computer vision library
• OpenCV Contrib - Additional OpenCV modules
• NVIDIA Video Codec SDK - Hardware-accelerated video encoding/decoding
• CUDA Toolkit - NVIDIA's parallel computing platform
• FFmpeg - Multimedia framework
• Cupy - GPU-accelerated NumPy implementation

License

This project is licensed under the MIT License - see the LICENSE.md file for details.

References

1. OpenCV Documentation: https://docs.opencv.org/
2. NVIDIA CUDA Documentation: https://docs.nvidia.com/cuda/
3. NVIDIA Video Codec SDK Documentation: https://developer.nvidia.com/nvidia-video-codec-sdk/documentation
4. FFmpeg Documentation: https://ffmpeg.org/documentation.html
5. Cupy Documentation: https://docs.cupy.dev/

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Acknowledgments

• OpenCV team for the excellent computer vision library
• NVIDIA for CUDA and Video Codec SDK
• FFmpeg project for video processing capabilities