This repository is for camera calibration in C++.

camera_calibration_pkg
    │  CMakeLists.txt
    │  package.xml
    │
    ├─config
    │      calibration_param.yaml
    │      distortion_gridpattern_compare1.png
    │      distortion_gridpattern_compare2.png
    │      setup_aruco.yaml
    │      setup_chessboard.yaml
    │      setup_ros.yaml
    │
    ├─include
    │      camera_calibration_tool.h
    │      drawing_tool.h
    │      param_code.h
    │      undistortion.h
    │
    ├─launch
    │      aruco_ext_calib.launch
    │      aruco_marker_generate.launch
    │      aruco_video_ext_calib.launch
    │      camera_calibration.launch
    │      chessboard_cam_ext_calibration.launch
    │      opencv_undistortion_time_cost_test.launch
    │      show_distorted_grid_pattern.launch
    │      undistortion.launch
    │
    ├─lib
    │      aruco_tool.cpp
    │      camera_calibration_tool.cpp
    │      drawing_tool.cpp
    │      undistortion.cpp
    │
    └─src
            aruco_ext_calib.cpp
            aruco_marker_generate.cpp
            aruco_video_ext_calib.cpp
            camera_calibration.cpp
            chessboard_cam_ext_calibration.cpp
            opencv_undistortion_time_cost_test.cpp
            show_distorted_grid_pattern.cpp
            undistortion_sample.cpp

To use these nodes, please use launch file for transmitting server parameters.

1. Use cv::getOptimalNewCameraMatrix() and cv::initUndistortRectifyMap() to get map1 and map2. Map1 contains a mapping table on (x, y) which refer to the source gridpattern image pixel, and map2 is a mapping table on y for refinement of the image. Map1 and map2 is a mapping from distorted image to undistorted image.

2. Find a mapping from undistorted image to distorted image for drawing a distorted gridpattern from a uniform grid pattern: Assume $\ X$ is the original image matrix, $\ X'$ is the target image processed by undistortion, $\ X''$ is the distorted gridpattern image we need to generate, and $\ T$ is the undistortion transform matrix. We have

   $\ X' = TX$
   $\ ΔX = X - X' = X - TX = (I - T)X$
   $\ X'' = X + ΔX = X + (I - T)X = (2I - T)X$

   Therefore, the new transform matrix is $\ G = 2I - T$

   Implementation

   newMap1.at<cv::Vec2f>(y, x) = static_cast<cv::Vec2f>(cv::Vec2f(2.0 * x, 2.0 * y) - map1.at<cv::Vec2f>(y, x));
   #This is for map1 CV_32FC2 format.
   
   newMap2.at<uchar>(y, x) = static_cast<uchar>(2 * y - map2.at<uchar>(y, x));
   #This is for map2 CV_8U format.
   
   #For other format, please check the format table for cv::Mat::at() method to edit it. You can use cv::Mat::type() function to find the format code. 
   

3. Remapping by cv::remap() function.

After the camera intrinsics calibrating, real-time external calibration can be performed using an Aruco Marker with known true dimensions. The resulting rotation and translation matrices enable mapping from 3d space to 2d space. A simple function of AR can be realized by inserting objects (pictures, point clouds, etc.) into the 3d space.

Please use launch command to run the nodes. Before use it, please edit the yaml file in ./config folder.

QuintinUmi