56 lines
2.5 KiB
TeX
56 lines
2.5 KiB
TeX
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
% %
|
|
% C++ %
|
|
% %
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
|
|
\ifCpp
|
|
\section{Camera calibration}
|
|
The goal of this tutorial is to learn how to calibrate a camera given a set of chessboard images.
|
|
|
|
\texttt{Test data}: use images in your data/chess folder.
|
|
|
|
Compile opencv with samples by setting BUILD\_EXAMPLES to ON in cmake configuration.
|
|
|
|
Go to bin folder and use \texttt{imagelist\_creator} to create an xml/yaml list of your images. Then, run \texttt{calibration} sample to get camera parameters. Use square size equal to 3cm.
|
|
|
|
\section{Pose estimation}
|
|
Now, let us write a code that detects a chessboard in a new image and finds its distance from the camera. You can apply the same method to any object with knwon 3d geometry that you can detect in an image.
|
|
|
|
\texttt{Test data}: use chess\_test*.jpg images from your data folder.
|
|
|
|
Create an empty console project. Load a test image:
|
|
\begin{lstlisting}
|
|
Mat img = imread(argv[1], CV_LOAD_IMAGE_GRAYSCALE);
|
|
\end{lstlisting}
|
|
|
|
Detect a chessboard in this image using findChessboard function.
|
|
\begin{lstlisting}
|
|
bool found = findChessboardCorners( img, boardSize, ptvec, CV_CALIB_CB_ADAPTIVE_THRESH );
|
|
\end{lstlisting}
|
|
|
|
Now, write a function that generates a \texttt{vector<Point3f>} array of 3d coordinates of a chessboard in any coordinate system. For simplicity, let us choose a system such that one of the chessboard corners is in the origin and the board is in the plane \(z = 0\).
|
|
|
|
Read camera parameters from xml/yaml file:
|
|
\begin{lstlisting}
|
|
FileStorage fs(filename, FileStorage::READ);
|
|
Mat intrinsics, distortion;
|
|
fs["camera_matrix"] >> intrinsics;
|
|
fs["distortion_coefficients"] >> distortion;
|
|
\end{lstlisting}
|
|
|
|
Now we are ready to find chessboard pose by running solvePnP:
|
|
\begin{lstlisting}
|
|
vector<Point3f> boardPoints;
|
|
// fill the array
|
|
...
|
|
|
|
solvePnP(Mat(boardPoints), Mat(foundBoardCorners), cameraMatrix,
|
|
distCoeffs, rvec, tvec, false);
|
|
\end{lstlisting}
|
|
|
|
Calculate reprojection error like it is done in \texttt{calibration} sample (see textttt{opencv/samples/cpp/calibration.cpp}, function \texttt{computeReprojectionErrors}).
|
|
|
|
How to calculate the distance from the camera origin to any of the corners?
|
|
\fi |