edgeMatching

Pyramid search matching based on edge gradient cosine similarity

This project provides functionality for template matching using OpenCV. It includes shared libraries for template creation and finding, as well as executables for training and inference.

The similarity is defined as:

Split into normalized vectors:

Then, the cosine similarity becomes:

Performance Tests (CPU I7-10700)

Test	Resolution	Metrics	Image
Test1	3648*3648	Execution Time: 207ms Inference Parameters: f_angle_start=-5 f_angle_stop=5 f_min_score=0.7 f_matche_numbers=200 f_max_overlap=0.5
Test2	4024*3036	Execution Time: 109ms Inference Parameters: f_angle_start=-180 f_angle_stop=180 f_min_score=0.7 f_matche_numbers=200 f_max_overlap=0.5
Test3	2448*2048	Execution Time: 64ms Inference Parameters: f_angle_start=-180 f_angle_stop=180 f_min_score=0.7 f_matche_numbers=200 f_max_overlap=0.5
Test4	2592*1944	Execution Time: 41ms Inference Parameters: f_angle_start=-180 f_angle_stop=180 f_min_score=0.7 f_matche_numbers=200 f_max_overlap=0.5
Test5	4096*3000	Execution Time: 160ms Inference Parameters: f_angle_start=-180 f_angle_stop=180 f_min_score=0.85 f_matche_numbers=200 f_max_overlap=0.5

Note

Some pictures are sourced from NCC

Project Structure

The project directory structure is as follows:

.
├── 3rdparty
│   └── opencv # OpenCV dependencies
│       ├── bin # Executable binaries
│       ├── include # OpenCV headers
│       ├── lib # OpenCV libraries
│       └── share # OpenCV share files
├── assert # Test images for matching
├── build_opencv_with_contrib.sh # Script to build OpenCV with contrib modules
├── CMakeLists.txt # CMake build configuration file
├── include # Project header files
│   ├── FindTemplateV1.h
│   ├── MakeTemplateV1.h
│   ├── ROI.h
│   ├── Timer.h
│   └── Type.h
├── inference.cpp # Inference executable source code
├── LICENSE
├── README.md
├── src # Source code for the libraries
│   ├── FindTemplateV1.cpp
│   └── MakeTemplateV1.cpp
└── train.cpp # Training executable source code

Prerequisites

Ensure that you have the following installed in the Linux:

• CMake (version 3.10 or higher)
• g++ compiler (with support for C++11)
• OpenCV 4.7.0 (with contrib modules, optionally)

Building OpenCV

To build OpenCV with contrib modules, use the provided build_opencv_with_contrib.sh script.

if you cmake OpenCV 4.7.0 with contrib modules failed, you can download .cache.zip file from the following link: https://wwyn.lanzout.com/iB7Mb34k8kwd and extract it to the your_opencv_source_dir/.cache directory.

Steps to build OpenCV:

1. Open a terminal and navigate to the project root directory.

2. Run the script:

   ./build_opencv_with_contrib.sh
   

This will download and build OpenCV4.7.0 along with the contrib modules, placing the necessary files under the 3rdparty/opencv directory.

Using CMake to Build the Project

1. Create a build directory:

   mkdir build
   cd build

2. Run CMake to configure the project:

   cmake ..

3. Build the project using make:

   make -j$(nproc)

Running the Executables

After building the project, you can run the executables:

• To run the training executable:

  ./train

  or

  ./train ../assert/m1.png

• To run the inference executable:

  ./inference

  or

  ./inference ../assert/src.bmp

Compiling with g++

If you prefer to compile using g++ directly instead of using CMake, you can compile the source code manually.

1. Compile the libraries:

   • For compiling the shared library libMakeTemplateV1.so

   g++ -std=c++11 -O3 -fopenmp -fPIC -march=native -msse -msse2 -msse3 -msse4 -mavx -o libMakeTemplateV1.so -shared src/MakeTemplateV1.cpp \
   -I./include \
   -I./3rdparty/opencv/include \
   -L./3rdparty/opencv/lib \
   -lopencv_core -lopencv_imgproc -lopencv_highgui -lopencv_imgcodecs -lopencv_calib3d \
   -pthread

   • For compiling the shared library libFindTemplateV1.so

   g++ -std=c++11 -O3 -fopenmp -fPIC -march=native -msse -msse2 -msse3 -msse4 -mavx -o libFindTemplateV1.so -shared src/FindTemplateV1.cpp \
   -I./include \
   -I./3rdparty/opencv/include \
   -L./3rdparty/opencv/lib \
   -lopencv_core -lopencv_imgproc -lopencv_highgui -lopencv_imgcodecs -lopencv_calib3d \
   -pthread

2. Compile the executables:

   • For the train executable

   g++ train.cpp src/MakeTemplateV1.cpp -std=c++11 -O3 -fopenmp -Iinclude -I3rdparty/opencv/include -I3rdparty/opencv/include/opencv4 -L3rdparty/opencv/lib -lopencv_core -lopencv_imgproc -lopencv_highgui -lopencv_imgcodecs -lopencv_features2d -lopencv_flann -lopencv_calib3d -lopencv_videoio -lopencv_ximgproc -lopencv_xfeatures2d -o train

   • For the inference executable

   g++ inference.cpp src/FindTemplateV1.cpp -std=c++11 -O3 -fopenmp -Iinclude -I3rdparty/opencv/include -I3rdparty/opencv/include/opencv4 -L3rdparty/opencv/lib -lopencv_core -lopencv_imgproc -lopencv_highgui -lopencv_imgcodecs -lopencv_features2d -lopencv_flann -lopencv_calib3d -lopencv_videoio -lopencv_ximgproc -lopencv_xfeatures2d -o inference

Running the Executables

After compilation, you can run the executables as follows:

• To run the training executable:

  ./train assert/m1.png

• To run the inference executable:

  ./inference assert/src.bmp

Install the Libraries and Header

To install the libraries and header files to the system directories:

1. Run the following command to install:

   make install

This will copy the shared libraries and headers to the appropriate system directories.

• The shared libraries will be installed in the publish/lib directory.
• The executables will be installed in the publish/bin directory.
• The header files will be installed in the publish/include/shapeMatch directory.

License

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