Markus Konrad [email protected], June 2014
INKA Research Group / Project MINERVA, HTW Berlin - http://inka.htw-berlin.de/inka/projekte/minerva/
This repository contains a simple OpenCV-based automatic camera calibration tool and a database containing XML files with camera matrix and distortion coefficients for different devices that have so far been determined using this tool.
A camera's matrix and distortion coefficients is especially important in Augmented Reality applications, where real a object's 3D pose needs to be approximated from 2D camera images.
For more information, see the excellent OpenCV tutorial on camera calibration.
OpenCV also contains source code for a camera calibration tool, which is also the inspiration for this tool (and parts of it's code have been directly used here, too). However, for most cases it is rather bloated and more difficult to use, because it requires to write a XML file with settings first. The presented tool here can be controlled completely with program arguments (see "Usage" below). It can also run "headless", meaning without graphical output, which speeds up the whole process.
The program can be called in the following way:
./cam-intrinsics-db [-(g|i)pazhv] <square size in meters> [device]
square size in metersis necessary and specifies the side length of the squares in the chessboard grid that was used for calibration- optional flags:
- 'g' for graphical output (shows original and undistorted first frame)
- 'i' for interactive graphical output (step through all frames)
- 'p' to fix principal point during calibration
- 'a' to fix aspect ratio during calibration
- 'z' to assume zero tangential distortion during calibration
- 'h' to flip image horizontally
- 'v' to flip image vertically
- optionally specify a 'device' for which calibration photos or videos exist in the 'device_data/' folder; if no device is specified, the program will calibrate all devices in the device_data folder (with the same square size setting)
Calibration data, which means images and/or videos with that depict a chessboard grid from different angles, resides for each device in the folder device_data. You can download and print out such a grid from the assets folder.
The database folder contains an XML file for each camera that was calibrated with this tool so far. The XML files use the opencv_storage format to store the following calibration information:
Camera_Matrixcontains the 3x3 camera matrixDistortion_Coefficientscontains the 5 distortion coefficientsAvg_Reprojection_Erroris the reported average reprojection error that resulted from the above calibration information
These XML files can be loaded and used in your (AR) program, like for example with OpenCV's FileStorage:
cv::FileStorage fs;
cv::Mat camMat;
cv::Mat distCoeff;
fs.open(path_to_xml_file, cv::FileStorage::READ);
if (fs.isOpened()) {
fs["Camera_Matrix"] >> camMat;
fs["Distortion_Coefficients"] >> distCoeff;
}
- OpenCV 2.x headers and library files need to be installed
This program can be compiled with any standard C++ compiler, like e.g. g++.
To compile you should at first make sure the settings for the header search path and library search path are correct. If necessary, edit the CFLAGS and LDFLAGS parameters in the Makefile.
Run make to compile the program.
You can run make clean to remove the object file and the executable.
If you have a camera that you calibrated with this tool, please submit your data set from the device_data folder (calibration images / video) and the resulting XML file. You can for example do so by sending a pull request.
Note that you can only use chessboard grids for calibration. You can download and print out such a grid from the assets folder.
Also note, that your calibration data set should be smaller than 50MiB.
- add the following as optional program arguments:
- board size
- number of frames to use from videos
See LICENSE file.