CudaFrameForge

This repository contains the project submission by Euisuh Jeong for the CUDA at Scale for the Enterprise course. The project focuses on implementing edge detection in a video, using Scharr as an example, which can be applied as part of a lane detection algorithm.

Input and Output Examples

Input	Output

You can view the full output video here, and the original input video here.

Project Overview

The program takes a video file as input and produces an output video file. It performs the following steps:

• Reads each frame of the input video using OpenCV.
• Extracts the green channel of each frame.
• Applies Scharr edge detection to detect edges.
• Generates a grayscale frame by copying the new green channel onto each RGB channel.
• Writes the grayscale frame to the output file.

For memory allocation, the program utilizes 2D memory functions, which offer improved performance especially when padding is necessary.

Building and Testing

1. Build: Use the command make build to build the program.

2. Run: Supply the input and output video file paths as arguments. Options include:

   • Execute ./run.sh.
   • Use make run ARGS="data/video.mp4 data/out.mp4".
   • Directly execute ./venturi_processor data/video.mp4 data/out.mp4.