\cvarg{useDisparityGuess}{If the parameter is not zero, the algorithm will start with pre-defined disparity maps. Both dispLeft and dispRight should be valid disparity maps. Otherwise, the function starts with blank disparity maps (all pixels are marked as occlusions).}
\end{description}
The function computes disparity maps for the input rectified stereo pair. Note that the left disparity image will contain values in the following range:
The function computes disparity maps for the input rectified stereo pair. Note that the left disparity image will contain values in the following range:
\cvarg{distCoeffs}{The input vector of distortion coefficients $(k_1, k_2, p_1, p_2[, k_3[, k_4, k_5, k_6]])$ of 4, 5 or 8 elements. If the vector is NULL/empty, the zero distortion coefficients are assumed.}
...
...
@@ -1064,7 +1064,7 @@ Finds the initial camera matrix from the 3D-2D point correspondences
where $(k_1, k_2, p_1, p_2[, k_3])$ are the distortion coefficients.
In the case of a stereo camera this function is called twice, once for each camera head, after \cvCross{StereoRectify}{stereoRectify}, which in its turn is called after \cvCross{StereoCalibrate}{stereoCalibrate}. But if the stereo camera was not calibrated, it is still possible to compute the rectification transformations directly from the fundamental matrix using \cvCross{StereoRectifyUncalibrated}{stereoRectifyUncalibrated}. For each camera the function computes homography \texttt{H} as the rectification transformation in pixel domain, not a rotation matrix \texttt{R} in 3D space. The \texttt{R} can be computed from \texttt{H} as
where $(k_1, k_2, p_1, p_2[, k_3])$ are the distortion coefficients.
In the case of a stereo camera this function is called twice, once for each camera head, after \cvCross{StereoRectify}{stereoRectify}, which in its turn is called after \cvCross{StereoCalibrate}{stereoCalibrate}. But if the stereo camera was not calibrated, it is still possible to compute the rectification transformations directly from the fundamental matrix using \cvCross{StereoRectifyUncalibrated}{stereoRectifyUncalibrated}. For each camera the function computes homography \texttt{H} as the rectification transformation in pixel domain, not a rotation matrix \texttt{R} in 3D space. The \texttt{R} can be computed from \texttt{H} as
@@ -1280,8 +1280,8 @@ Reprojects disparity image to 3D space.
\cvarg{Q}{The $4\times4$ perspective transformation matrix that can be obtained with \cvCross{StereoRectify}{stereoRectify}}
\cvarg{handleMissingValues}{If true, when the pixels with the minimal disparity (that corresponds to the outliers; see \cvCross{FindStereoCorrespondenceBM}{StereoBM}) will be transformed to 3D points with some very large Z value (currently set to 10000)}
\end{description}
The function transforms 1-channel disparity map to 3-channel image representing a 3D surface. That is, for each pixel \texttt{(x,y)} and the corresponding disparity \texttt{d=disparity(x,y)} it computes:
The function transforms 1-channel disparity map to 3-channel image representing a 3D surface. That is, for each pixel \texttt{(x,y)} and the corresponding disparity \texttt{d=disparity(x,y)} it computes:
\cvarg{roi1, roi2}{The optional output rectangles inside the rectified images where all the pixels are valid. If \texttt{alpha=0}, the ROIs will cover the whole images, otherwise they likely be smaller, see the picture below}
\end{description}
The function computes the rotation matrices for each camera that (virtually) make both camera image planes the same plane. Consequently, that makes all the epipolar lines parallel and thus simplifies the dense stereo correspondence problem. On input the function takes the matrices computed by \cvCppCross{stereoCalibrate} and on output it gives 2 rotation matrices and also 2 projection matrices in the new coordinates. The 2 cases are distinguished by the function are:
The function computes the rotation matrices for each camera that (virtually) make both camera image planes the same plane. Consequently, that makes all the epipolar lines parallel and thus simplifies the dense stereo correspondence problem. On input the function takes the matrices computed by \cvCppCross{stereoCalibrate} and on output it gives 2 rotation matrices and also 2 projection matrices in the new coordinates. The 2 cases are distinguished by the function are:
\begin{enumerate}
\item Horizontal stereo, when 1st and 2nd camera views are shifted relative to each other mainly along the x axis (with possible small vertical shift). Then in the rectified images the corresponding epipolar lines in left and right cameras will be horizontal and have the same y-coordinate. P1 and P2 will look as:
\item Horizontal stereo, when 1st and 2nd camera views are shifted relative to each other mainly along the x axis (with possible small vertical shift). Then in the rectified images the corresponding epipolar lines in left and right cameras will be horizontal and have the same y-coordinate. P1 and P2 will look as:
\[\texttt{P1}=
\begin{bmatrix}
...
...
@@ -1711,9 +1711,9 @@ f & 0 & cx & 0\\
\]
where $T_y$ is vertical shift between the cameras and $cy_1=cy_2$ if \texttt{CALIB\_ZERO\_DISPARITY} is set.
\end{enumerate}
\end{enumerate}
As you can see, the first 3 columns of \texttt{P1} and \texttt{P2} will effectively be the new "rectified" camera matrices.
As you can see, the first 3 columns of \texttt{P1} and \texttt{P2} will effectively be the new "rectified" camera matrices.
The matrices, together with \texttt{R1} and \texttt{R2}, can then be passed to \cvCross{InitUndistortRectifyMap}{initUndistortRectifyMap} to initialize the rectification map for each camera.
Below is the screenshot from \texttt{stereo\_calib.cpp} sample. Some red horizontal lines, as you can see, pass through the corresponding image regions, i.e. the images are well rectified (which is what most stereo correspondence algorithms rely on). The green rectangles are \texttt{roi1} and \texttt{roi2} - indeed, their interior are all valid pixels.
The function computes the rectification transformations without knowing intrinsic parameters of the cameras and their relative position in space, hence the suffix "Uncalibrated". Another related difference from \cvCross{StereoRectify}{stereoRectify} is that the function outputs not the rectification transformations in the object (3D) space, but the planar perspective transformations, encoded by the homography matrices \texttt{H1} and \texttt{H2}. The function implements the algorithm \cite{Hartley99}.
The function computes the rectification transformations without knowing intrinsic parameters of the cameras and their relative position in space, hence the suffix "Uncalibrated". Another related difference from \cvCross{StereoRectify}{stereoRectify} is that the function outputs not the rectification transformations in the object (3D) space, but the planar perspective transformations, encoded by the homography matrices \texttt{H1} and \texttt{H2}. The function implements the algorithm \cite{Hartley99}.
Note that while the algorithm does not need to know the intrinsic parameters of the cameras, it heavily depends on the epipolar geometry. Therefore, if the camera lenses have significant distortion, it would better be corrected before computing the fundamental matrix and calling this function. For example, distortion coefficients can be estimated for each head of stereo camera separately by using \cvCross{CalibrateCamera2}{calibrateCamera} and then the images can be corrected using \cvCross{Undistort2}{undistort}, or just the point coordinates can be corrected with \cvCross{UndistortPoints}{undistortPoints}.
\cvarg{src}{The observed point coordinates, 1xN or Nx1 2-channel (CV\_32FC2 or CV\_64FC2).}
\cvarg{src}{The observed point coordinates, 1xN or Nx1 2-channel (CV\_32FC2 or CV\_64FC2).}
\cvarg{dst}{The output ideal point coordinates, after undistortion and reverse perspective transformation\cvCPy{, same format as \texttt{src}}.}
\cvarg{cameraMatrix}{The camera matrix $\vecthreethree{f_x}{0}{c_x}{0}{f_y}{c_y}{0}{0}{1}$}
\cvarg{distCoeffs}\cvarg{distCoeffs}{The input vector of distortion coefficients $(k_1, k_2, p_1, p_2[, k_3[, k_4, k_5, k_6]])$ of 4, 5 or 8 elements. If the vector is NULL/empty, the zero distortion coefficients are assumed.}
int featureType; // feature type (HAAR or LBP for now)
int ncategories; // number of categories (for categorical features only)
int ncategories; // number of categories (for categorical features only)
Size origWinSize; // size of training images
vector<Stage> stages; // vector of stages (BOOST for now)
vector<DTree> classifiers; // vector of decision trees
vector<DTreeNode> nodes; // vector of tree nodes
...
...
@@ -635,5 +635,150 @@ Groups the object candidate rectangles
\cvarg{eps}{The relative difference between sides of the rectangles to merge them into a group}
\end{description}
The function is a wrapper for a generic function \cvCppCross{partition}. It clusters all the input rectangles using the rectangle equivalence criteria, that combines rectangles that have similar sizes and similar locations (the similarity is defined by \texttt{eps}). When \texttt{eps=0}, no clustering is done at all. If $\texttt{eps}\rightarrow+\inf$, all the rectangles will be put in one cluster. Then, the small clusters, containing less than or equal to \texttt{groupThreshold} rectangles, will be rejected. In each other cluster the average rectangle will be computed and put into the output rectangle list.
The function is a wrapper for a generic function \cvCppCross{partition}. It clusters all the input rectangles using the rectangle equivalence criteria, that combines rectangles that have similar sizes and similar locations (the similarity is defined by \texttt{eps}). When \texttt{eps=0}, no clustering is done at all. If $\texttt{eps}\rightarrow+\inf$, all the rectangles will be put in one cluster. Then, the small clusters, containing less than or equal to \texttt{groupThreshold} rectangles, will be rejected. In each other cluster the average rectangle will be computed and put into the output rectangle list.
\fi
\ifC
\section{Discriminatively Trained Part Based Models for Object Detection}
\subsection{Discriminatively Trained Part Based Models for Object Detection}
The object detector described below has been initially proposed by
P.F. Felzenszwalb in \cvCPyCross{Felzenszwalb10}. It is based on a
Dalal-Triggs detector that uses a single filter on histogram of
oriented gradients (HOG) features to represent an object category.
This detector uses a sliding window approach, where a filter is
applied at all positions and scales of an image. The first
innovation is enriching the Dalal-Triggs model using a
star-structured part-based model defined by a "root" filter
(analogous to the Dalal-Triggs filter) plus a set of parts filters
and associated deformation models. The score of one of star models
at a particular position and scale within an image is the score of
the root filter at the given location plus the sum over parts of the
maximum, over placements of that part, of the part filter score on
its location minus a deformation cost measuring the deviation of the
part from its ideal location relative to the root. Both root and
part filter scores are defined by the dot product between a filter
(a set of weights) and a subwindow of a feature pyramid computed
from the input image. Another improvement is a representation of the
class of models by a mixture of star models. The score of a mixture
model at a particular position and scale is the maximum over
components, of the score of that component model at the given
title = {Theory and Practice of Projective Rectification},
...
...
@@ -311,4 +323,3 @@
# '''[Zhang96]''' Z. Zhang. Parameter Estimation Techniques: A Tutorial with Application to Conic Fitting, Image and Vision Computing Journal, 1996.
# '''[Zhang99]''' Z. Zhang. Flexible Camera Calibration By Viewing a Plane From Unknown Orientations. International Conference on Computer Vision (ICCV'99), Corfu, Greece, pages 666-673, September 1999.
# '''[Zhang00]''' Z. Zhang. A Flexible New Technique for Camera Calibration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(11):1330-1334, 2000.
