CUDA-Harris-Corner-Detector

This project is a CUDA-acclerated Harris Corner Detector based on Corner-Detector.
We implemented the project using C/C++ with i7-7700 and RTX 2080Ti.

Usage

Folder and Data

Here is the directory tree structure for the project.

├── input
│   └── image.ppm           // input images(should be ppm format)
├── Makefile
├── output
│   └── ...                 // output images
├── README.md
└── source
    ├── CornerDetector.cu   // main part of the project
    ├── GaussFilter.cuh     // CUDA header file of gaussian filter 
    ├── Gauss.h             // header file of gaussian filter
    ├── Matrix.h            // header file of class Matrix
    ├── PPM.h               // header file to deal with portable pixmap format (PPM) image
    ├── SobelFilter.cuh     // CUDA header file of sobel filter 
    ├── Sobel.h             // header file of sobel filter
    ├── utils.h             // other utilities
    └── VectorOperation.cuh // CUDA header file of vector operation

Prepare Images

Put your images into the ./input folder. Please note that the image must be .ppm format. With linux you can convert the image format using

convert image.jpg image.ppm

Start

here is an example to run the code.

Compilation

you can just use make command, or

nvcc CornerDetector.cu -o CornerDetector

Execution

There are four parameter to assign.

./CornerDetector <imagePath> <gaussMask=size> <tpb=threadsPerBlock> <sigma=doubleValue>

example

./CornerDetector ./input/IMG_0125.ppm

Results

To compare the perfomance between CPU and GPU, we perform our Harris Corner Detector on a image of size 3648x5472, and the comparison results shown in Table 1. We could see that the GPU version is much faster the CPU version.

The output image shown below. The red spot are Harris Corner Response.

Result from IMG_0125.ppm	Result from IMG_4486.ppm

Reference

• Digital Visual Effects, Yung-Yu Chuang
• Introduction to Cuda Parallel Programming, Ting-Wai Chiu
• Corner-Detector, jariasf