Custom Calibration Pattern for 3D reconstruction

Detailed Description

Namespaces
namespace	cv::omnidir::internal

Classes
class	cv::ccalib::CustomPattern
class	cv::multicalib::MultiCameraCalibration
	Class for multiple camera calibration that supports pinhole camera and omnidirection camera. For omnidirectional camera model, please refer to omnidir.hpp in ccalib module. It first calibrate each camera individually, then a bundle adjustment like optimization is applied to refine extrinsic parameters. So far, it only support "random" pattern for calibration, see randomPattern.hpp in ccalib module for details. Images that are used should be named by "cameraIdx-timestamp.*", several images with the same timestamp means that they are the same pattern that are photographed. cameraIdx should start from 0. More...
class	cv::randpattern::RandomPatternCornerFinder
	Class for finding features points and corresponding 3D in world coordinate of a "random" pattern, which can be to be used in calibration. It is useful when pattern is partly occluded or only a part of pattern can be observed in multiple cameras calibration. The pattern can be generated by RandomPatternGenerator class described in this file. More...
class	cv::randpattern::RandomPatternGenerator

Macros
#define	HEAD   -1
#define	INVALID   -2

Enumerations
enum	{   cv::omnidir::CALIB_USE_GUESS = 1 ,   cv::omnidir::CALIB_FIX_SKEW = 2 ,   cv::omnidir::CALIB_FIX_K1 = 4 ,   cv::omnidir::CALIB_FIX_K2 = 8 ,   cv::omnidir::CALIB_FIX_P1 = 16 ,   cv::omnidir::CALIB_FIX_P2 = 32 ,   cv::omnidir::CALIB_FIX_XI = 64 ,   cv::omnidir::CALIB_FIX_GAMMA = 128 ,   cv::omnidir::CALIB_FIX_CENTER = 256 }
enum	{   cv::omnidir::RECTIFY_PERSPECTIVE = 1 ,   cv::omnidir::RECTIFY_CYLINDRICAL = 2 ,   cv::omnidir::RECTIFY_LONGLATI = 3 ,   cv::omnidir::RECTIFY_STEREOGRAPHIC = 4 }
enum	{   cv::omnidir::XYZRGB = 1 ,   cv::omnidir::XYZ = 2 }

Functions
double	cv::omnidir::calibrate (InputArrayOfArrays objectPoints, InputArrayOfArrays imagePoints, Size size, InputOutputArray K, InputOutputArray xi, InputOutputArray D, OutputArrayOfArrays rvecs, OutputArrayOfArrays tvecs, int flags, TermCriteria criteria, OutputArray idx=noArray())
	Perform omnidirectional camera calibration, the default depth of outputs is CV_64F.
void	cv::omnidir::initUndistortRectifyMap (InputArray K, InputArray D, InputArray xi, InputArray R, InputArray P, const cv::Size &size, int m1type, OutputArray map1, OutputArray map2, int flags)
	Computes undistortion and rectification maps for omnidirectional camera image transform by a rotation R. It output two maps that are used for cv::remap(). If D is empty then zero distortion is used, if R or P is empty then identity matrices are used.
void	cv::omnidir::projectPoints (InputArray objectPoints, OutputArray imagePoints, const Affine3d &affine, InputArray K, double xi, InputArray D, OutputArray jacobian=noArray())
void	cv::omnidir::projectPoints (InputArray objectPoints, OutputArray imagePoints, InputArray rvec, InputArray tvec, InputArray K, double xi, InputArray D, OutputArray jacobian=noArray())
	Projects points for omnidirectional camera using CMei's model.
double	cv::omnidir::stereoCalibrate (InputOutputArrayOfArrays objectPoints, InputOutputArrayOfArrays imagePoints1, InputOutputArrayOfArrays imagePoints2, const Size &imageSize1, const Size &imageSize2, InputOutputArray K1, InputOutputArray xi1, InputOutputArray D1, InputOutputArray K2, InputOutputArray xi2, InputOutputArray D2, OutputArray rvec, OutputArray tvec, OutputArrayOfArrays rvecsL, OutputArrayOfArrays tvecsL, int flags, TermCriteria criteria, OutputArray idx=noArray())
	Stereo calibration for omnidirectional camera model. It computes the intrinsic parameters for two cameras and the extrinsic parameters between two cameras. The default depth of outputs is CV_64F.
void	cv::omnidir::stereoReconstruct (InputArray image1, InputArray image2, InputArray K1, InputArray D1, InputArray xi1, InputArray K2, InputArray D2, InputArray xi2, InputArray R, InputArray T, int flag, int numDisparities, int SADWindowSize, OutputArray disparity, OutputArray image1Rec, OutputArray image2Rec, const Size &newSize=Size(), InputArray Knew=cv::noArray(), OutputArray pointCloud=cv::noArray(), int pointType=XYZRGB)
	Stereo 3D reconstruction from a pair of images.
void	cv::omnidir::stereoRectify (InputArray R, InputArray T, OutputArray R1, OutputArray R2)
	Stereo rectification for omnidirectional camera model. It computes the rectification rotations for two cameras.
void	cv::omnidir::undistortImage (InputArray distorted, OutputArray undistorted, InputArray K, InputArray D, InputArray xi, int flags, InputArray Knew=cv::noArray(), const Size &new_size=Size(), InputArray R=Mat::eye(3, 3, CV_64F))
	Undistort omnidirectional images to perspective images.
void	cv::omnidir::undistortPoints (InputArray distorted, OutputArray undistorted, InputArray K, InputArray D, InputArray xi, InputArray R)
	Undistort 2D image points for omnidirectional camera using CMei's model.

Macro Definition Documentation

HEAD

#define HEAD   -1

#include <opencv2/ccalib/multicalib.hpp>

INVALID

#define INVALID   -2

#include <opencv2/ccalib/multicalib.hpp>

Enumeration Type Documentation

anonymous enum

anonymous enum

#include <opencv2/ccalib/omnidir.hpp>

Enumerator
CALIB_USE_GUESS  Python: cv.omnidir.CALIB_USE_GUESS
CALIB_FIX_SKEW  Python: cv.omnidir.CALIB_FIX_SKEW
CALIB_FIX_K1  Python: cv.omnidir.CALIB_FIX_K1
CALIB_FIX_K2  Python: cv.omnidir.CALIB_FIX_K2
CALIB_FIX_P1  Python: cv.omnidir.CALIB_FIX_P1
CALIB_FIX_P2  Python: cv.omnidir.CALIB_FIX_P2
CALIB_FIX_XI  Python: cv.omnidir.CALIB_FIX_XI
CALIB_FIX_GAMMA  Python: cv.omnidir.CALIB_FIX_GAMMA
CALIB_FIX_CENTER  Python: cv.omnidir.CALIB_FIX_CENTER

anonymous enum

anonymous enum

#include <opencv2/ccalib/omnidir.hpp>

Enumerator
RECTIFY_PERSPECTIVE  Python: cv.omnidir.RECTIFY_PERSPECTIVE
RECTIFY_CYLINDRICAL  Python: cv.omnidir.RECTIFY_CYLINDRICAL
RECTIFY_LONGLATI  Python: cv.omnidir.RECTIFY_LONGLATI
RECTIFY_STEREOGRAPHIC  Python: cv.omnidir.RECTIFY_STEREOGRAPHIC

anonymous enum

anonymous enum

#include <opencv2/ccalib/omnidir.hpp>

Enumerator
XYZRGB  Python: cv.omnidir.XYZRGB
XYZ  Python: cv.omnidir.XYZ

Function Documentation

calibrate()

double cv::omnidir::calibrate	(	InputArrayOfArrays	objectPoints,
		InputArrayOfArrays	imagePoints,
		Size	size,
		InputOutputArray	K,
		InputOutputArray	xi,
		InputOutputArray	D,
		OutputArrayOfArrays	rvecs,
		OutputArrayOfArrays	tvecs,
		int	flags,
		TermCriteria	criteria,
		OutputArray	idx = noArray()
	)

Python:
	cv.omnidir.calibrate(	objectPoints, imagePoints, size, K, xi, D, flags, criteria[, rvecs[, tvecs[, idx]]]	) ->	retval, K, xi, D, rvecs, tvecs, idx

#include <opencv2/ccalib/omnidir.hpp>

Perform omnidirectional camera calibration, the default depth of outputs is CV_64F.

Parameters

objectPoints	Vector of vector of Vec3f object points in world (pattern) coordinate. It also can be vector of Mat with size 1xN/Nx1 and type CV_32FC3. Data with depth of 64_F is also acceptable.
imagePoints	Vector of vector of Vec2f corresponding image points of objectPoints. It must be the same size and the same type with objectPoints.
size	Image size of calibration images.
K	Output calibrated camera matrix.
xi	Output parameter xi for CMei's model
D	Output distortion parameters \((k_1, k_2, p_1, p_2)\)
rvecs	Output rotations for each calibration images
tvecs	Output translation for each calibration images
flags	The flags that control calibrate
criteria	Termination criteria for optimization
idx	Indices of images that pass initialization, which are really used in calibration. So the size of rvecs is the same as idx.total().

initUndistortRectifyMap()

void cv::omnidir::initUndistortRectifyMap	(	InputArray	K,
		InputArray	D,
		InputArray	xi,
		InputArray	R,
		InputArray	P,
		const cv::Size &	size,
		int	m1type,
		OutputArray	map1,
		OutputArray	map2,
		int	flags
	)

Python:
	cv.omnidir.initUndistortRectifyMap(	K, D, xi, R, P, size, m1type, flags[, map1[, map2]]	) ->	map1, map2

#include <opencv2/ccalib/omnidir.hpp>

Computes undistortion and rectification maps for omnidirectional camera image transform by a rotation R. It output two maps that are used for cv::remap(). If D is empty then zero distortion is used, if R or P is empty then identity matrices are used.

Parameters

K	Camera matrix \(K = \vecthreethree{f_x}{s}{c_x}{0}{f_y}{c_y}{0}{0}{_1}\), with depth CV_32F or CV_64F
D	Input vector of distortion coefficients \((k_1, k_2, p_1, p_2)\), with depth CV_32F or CV_64F
xi	The parameter xi for CMei's model
R	Rotation transform between the original and object space : 3x3 1-channel, or vector: 3x1/1x3, with depth CV_32F or CV_64F
P	New camera matrix (3x3) or new projection matrix (3x4)
size	Undistorted image size.
m1type	Type of the first output map that can be CV_32FC1 or CV_16SC2 . See convertMaps() for details.
map1	The first output map.
map2	The second output map.
flags	Flags indicates the rectification type, RECTIFY_PERSPECTIVE, RECTIFY_CYLINDRICAL, RECTIFY_LONGLATI and RECTIFY_STEREOGRAPHIC are supported.

projectPoints() [1/2]

void cv::omnidir::projectPoints	(	InputArray	objectPoints,
		OutputArray	imagePoints,
		const Affine3d &	affine,
		InputArray	K,
		double	xi,
		InputArray	D,
		OutputArray	jacobian = noArray()
	)

Python:
	cv.omnidir.projectPoints(	objectPoints, rvec, tvec, K, xi, D[, imagePoints[, jacobian]]	) ->	imagePoints, jacobian

#include <opencv2/ccalib/omnidir.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

projectPoints() [2/2]

void cv::omnidir::projectPoints	(	InputArray	objectPoints,
		OutputArray	imagePoints,
		InputArray	rvec,
		InputArray	tvec,
		InputArray	K,
		double	xi,
		InputArray	D,
		OutputArray	jacobian = noArray()
	)

Python:
	cv.omnidir.projectPoints(	objectPoints, rvec, tvec, K, xi, D[, imagePoints[, jacobian]]	) ->	imagePoints, jacobian

#include <opencv2/ccalib/omnidir.hpp>

Projects points for omnidirectional camera using CMei's model.

This module was accepted as a GSoC 2015 project for OpenCV, authored by Baisheng Lai, mentored by Bo Li.

Parameters

objectPoints	Object points in world coordinate, vector of vector of Vec3f or Mat of 1xN/Nx1 3-channel of type CV_32F and N is the number of points. 64F is also acceptable.
imagePoints	Output array of image points, vector of vector of Vec2f or 1xN/Nx1 2-channel of type CV_32F. 64F is also acceptable.
rvec	vector of rotation between world coordinate and camera coordinate, i.e., om
tvec	vector of translation between pattern coordinate and camera coordinate
K	Camera matrix \(K = \vecthreethree{f_x}{s}{c_x}{0}{f_y}{c_y}{0}{0}{_1}\).
D	Input vector of distortion coefficients \((k_1, k_2, p_1, p_2)\).
xi	The parameter xi for CMei's model
jacobian	Optional output 2Nx16 of type CV_64F jacobian matrix, contains the derivatives of image pixel points wrt parameters including \(om, T, f_x, f_y, s, c_x, c_y, xi, k_1, k_2, p_1, p_2\). This matrix will be used in calibration by optimization.

The function projects object 3D points of world coordinate to image pixels, parameter by intrinsic and extrinsic parameters. Also, it optionally compute a by-product: the jacobian matrix containing contains the derivatives of image pixel points wrt intrinsic and extrinsic parameters.

stereoCalibrate()

double cv::omnidir::stereoCalibrate	(	InputOutputArrayOfArrays	objectPoints,
		InputOutputArrayOfArrays	imagePoints1,
		InputOutputArrayOfArrays	imagePoints2,
		const Size &	imageSize1,
		const Size &	imageSize2,
		InputOutputArray	K1,
		InputOutputArray	xi1,
		InputOutputArray	D1,
		InputOutputArray	K2,
		InputOutputArray	xi2,
		InputOutputArray	D2,
		OutputArray	rvec,
		OutputArray	tvec,
		OutputArrayOfArrays	rvecsL,
		OutputArrayOfArrays	tvecsL,
		int	flags,
		TermCriteria	criteria,
		OutputArray	idx = noArray()
	)

Python:
	cv.omnidir.stereoCalibrate(	objectPoints, imagePoints1, imagePoints2, imageSize1, imageSize2, K1, xi1, D1, K2, xi2, D2, flags, criteria[, rvec[, tvec[, rvecsL[, tvecsL[, idx]]]]]	) ->	retval, objectPoints, imagePoints1, imagePoints2, K1, xi1, D1, K2, xi2, D2, rvec, tvec, rvecsL, tvecsL, idx

#include <opencv2/ccalib/omnidir.hpp>

Stereo calibration for omnidirectional camera model. It computes the intrinsic parameters for two cameras and the extrinsic parameters between two cameras. The default depth of outputs is CV_64F.

Parameters

objectPoints	Object points in world (pattern) coordinate. Its type is vector<vector<Vec3f> >. It also can be vector of Mat with size 1xN/Nx1 and type CV_32FC3. Data with depth of 64_F is also acceptable.
imagePoints1	The corresponding image points of the first camera, with type vector<vector<Vec2f> >. It must be the same size and the same type as objectPoints.
imagePoints2	The corresponding image points of the second camera, with type vector<vector<Vec2f> >. It must be the same size and the same type as objectPoints.
imageSize1	Image size of calibration images of the first camera.
imageSize2	Image size of calibration images of the second camera.
K1	Output camera matrix for the first camera.
xi1	Output parameter xi of Mei's model for the first camera
D1	Output distortion parameters \((k_1, k_2, p_1, p_2)\) for the first camera
K2	Output camera matrix for the first camera.
xi2	Output parameter xi of CMei's model for the second camera
D2	Output distortion parameters \((k_1, k_2, p_1, p_2)\) for the second camera
rvec	Output rotation between the first and second camera
tvec	Output translation between the first and second camera
rvecsL	Output rotation for each image of the first camera
tvecsL	Output translation for each image of the first camera
flags	The flags that control stereoCalibrate
criteria	Termination criteria for optimization
idx	Indices of image pairs that pass initialization, which are really used in calibration. So the size of rvecs is the same as idx.total(). @

stereoReconstruct()

void cv::omnidir::stereoReconstruct	(	InputArray	image1,
		InputArray	image2,
		InputArray	K1,
		InputArray	D1,
		InputArray	xi1,
		InputArray	K2,
		InputArray	D2,
		InputArray	xi2,
		InputArray	R,
		InputArray	T,
		int	flag,
		int	numDisparities,
		int	SADWindowSize,
		OutputArray	disparity,
		OutputArray	image1Rec,
		OutputArray	image2Rec,
		const Size &	newSize = Size(),
		InputArray	Knew = cv::noArray(),
		OutputArray	pointCloud = cv::noArray(),
		int	pointType = XYZRGB
	)

Python:
	cv.omnidir.stereoReconstruct(	image1, image2, K1, D1, xi1, K2, D2, xi2, R, T, flag, numDisparities, SADWindowSize[, disparity[, image1Rec[, image2Rec[, newSize[, Knew[, pointCloud[, pointType]]]]]]]	) ->	disparity, image1Rec, image2Rec, pointCloud

#include <opencv2/ccalib/omnidir.hpp>

Stereo 3D reconstruction from a pair of images.

Parameters

image1	The first input image
image2	The second input image
K1	Input camera matrix of the first camera
D1	Input distortion parameters \((k_1, k_2, p_1, p_2)\) for the first camera
xi1	Input parameter xi for the first camera for CMei's model
K2	Input camera matrix of the second camera
D2	Input distortion parameters \((k_1, k_2, p_1, p_2)\) for the second camera
xi2	Input parameter xi for the second camera for CMei's model
R	Rotation between the first and second camera
T	Translation between the first and second camera
flag	Flag of rectification type, RECTIFY_PERSPECTIVE or RECTIFY_LONGLATI
numDisparities	The parameter 'numDisparities' in StereoSGBM, see StereoSGBM for details.
SADWindowSize	The parameter 'SADWindowSize' in StereoSGBM, see StereoSGBM for details.
disparity	Disparity map generated by stereo matching
image1Rec	Rectified image of the first image
image2Rec	rectified image of the second image
newSize	Image size of rectified image, see omnidir::undistortImage
Knew	New camera matrix of rectified image, see omnidir::undistortImage
pointCloud	Point cloud of 3D reconstruction, with type CV_64FC3
pointType	Point cloud type, it can be XYZRGB or XYZ

stereoRectify()

void cv::omnidir::stereoRectify	(	InputArray	R,
		InputArray	T,
		OutputArray	R1,
		OutputArray	R2
	)

Python:
	cv.omnidir.stereoRectify(	R, T[, R1[, R2]]	) ->	R1, R2

#include <opencv2/ccalib/omnidir.hpp>

Stereo rectification for omnidirectional camera model. It computes the rectification rotations for two cameras.

Parameters

R	Rotation between the first and second camera
T	Translation between the first and second camera
R1	Output 3x3 rotation matrix for the first camera
R2	Output 3x3 rotation matrix for the second camera

undistortImage()

void cv::omnidir::undistortImage	(	InputArray	distorted,
		OutputArray	undistorted,
		InputArray	K,
		InputArray	D,
		InputArray	xi,
		int	flags,
		InputArray	Knew = cv::noArray(),
		const Size &	new_size = Size(),
		InputArray	R = Mat::eye(3, 3, CV_64F)
	)

Python:
	cv.omnidir.undistortImage(	distorted, K, D, xi, flags[, undistorted[, Knew[, new_size[, R]]]]	) ->	undistorted

#include <opencv2/ccalib/omnidir.hpp>

Undistort omnidirectional images to perspective images.

Parameters

distorted	The input omnidirectional image.
undistorted	The output undistorted image.
K	Camera matrix \(K = \vecthreethree{f_x}{s}{c_x}{0}{f_y}{c_y}{0}{0}{_1}\).
D	Input vector of distortion coefficients \((k_1, k_2, p_1, p_2)\).
xi	The parameter xi for CMei's model.
flags	Flags indicates the rectification type, RECTIFY_PERSPECTIVE, RECTIFY_CYLINDRICAL, RECTIFY_LONGLATI and RECTIFY_STEREOGRAPHIC
Knew	Camera matrix of the distorted image. If it is not assigned, it is just K.
new_size	The new image size. By default, it is the size of distorted.
R	Rotation matrix between the input and output images. By default, it is identity matrix.

undistortPoints()

void cv::omnidir::undistortPoints	(	InputArray	distorted,
		OutputArray	undistorted,
		InputArray	K,
		InputArray	D,
		InputArray	xi,
		InputArray	R
	)

Python:
	cv.omnidir.undistortPoints(	distorted, K, D, xi, R[, undistorted]	) ->	undistorted

#include <opencv2/ccalib/omnidir.hpp>

Undistort 2D image points for omnidirectional camera using CMei's model.

Parameters

distorted	Array of distorted image points, vector of Vec2f or 1xN/Nx1 2-channel Mat of type CV_32F, 64F depth is also acceptable
K	Camera matrix \(K = \vecthreethree{f_x}{s}{c_x}{0}{f_y}{c_y}{0}{0}{_1}\).
D	Distortion coefficients \((k_1, k_2, p_1, p_2)\).
xi	The parameter xi for CMei's model
R	Rotation trainsform between the original and object space : 3x3 1-channel, or vector: 3x1/1x3 1-channel or 1x1 3-channel
undistorted	array of normalized object points, vector of Vec2f/Vec2d or 1xN/Nx1 2-channel Mat with the same depth of distorted points.