![]() |
Visual Servoing Platform
version 3.3.0
|
#include <vpPose.h>
Public Types | |
enum | vpPoseMethodType { LAGRANGE, DEMENTHON, LOWE, RANSAC, LAGRANGE_LOWE, DEMENTHON_LOWE, VIRTUAL_VS, DEMENTHON_VIRTUAL_VS, LAGRANGE_VIRTUAL_VS } |
enum | RANSAC_FILTER_FLAGS { NO_FILTER, PREFILTER_DEGENERATE_POINTS, CHECK_DEGENERATE_POINTS } |
Static Public Member Functions | |
static void | display (vpImage< unsigned char > &I, vpHomogeneousMatrix &cMo, vpCameraParameters &cam, double size, vpColor col=vpColor::none) |
static void | display (vpImage< vpRGBa > &I, vpHomogeneousMatrix &cMo, vpCameraParameters &cam, double size, vpColor col=vpColor::none) |
static double | poseFromRectangle (vpPoint &p1, vpPoint &p2, vpPoint &p3, vpPoint &p4, double lx, vpCameraParameters &cam, vpHomogeneousMatrix &cMo) |
static int | computeRansacIterations (double probability, double epsilon, const int sampleSize=4, int maxIterations=2000) |
static void | findMatch (std::vector< vpPoint > &p2D, std::vector< vpPoint > &p3D, const unsigned int &numberOfInlierToReachAConsensus, const double &threshold, unsigned int &ninliers, std::vector< vpPoint > &listInliers, vpHomogeneousMatrix &cMo, const int &maxNbTrials=10000, bool useParallelRansac=true, unsigned int nthreads=0, bool(*func)(const vpHomogeneousMatrix &)=NULL) |
static bool | computePlanarObjectPoseFromRGBD (const vpImage< float > &depthMap, const std::vector< vpImagePoint > &corners, const vpCameraParameters &colorIntrinsics, const std::vector< vpPoint > &point3d, vpHomogeneousMatrix &cMo, double *confidence_index=NULL) |
Public Attributes | |
unsigned int | npt |
std::list< vpPoint > | listP |
double | residual |
Protected Member Functions | |
double | computeResidualDementhon (const vpHomogeneousMatrix &cMo) |
int | calculArbreDementhon (vpMatrix &b, vpColVector &U, vpHomogeneousMatrix &cMo) |
Protected Attributes | |
double | lambda |
Class used for pose computation from N points (pose from point only). Some of the algorithms implemented in this class are described in [Marchand16a].
To see how to use this class you can follow the Tutorial: Pose estimation from points.
Methods that could be used to estimate the pose from points.
vpPose::vpPose | ( | ) |
Default constructor.
Definition at line 95 of file vpPose.cpp.
vpPose::vpPose | ( | const std::vector< vpPoint > & | lP | ) |
Definition at line 105 of file vpPose.cpp.
|
virtual |
Destructor that deletes the array of point (freed the memory).
Definition at line 118 of file vpPose.cpp.
void vpPose::addPoint | ( | const vpPoint & | newP | ) |
Add a new point in the array of points.
newP | : New point to add in the array of point. |
Definition at line 148 of file vpPose.cpp.
Referenced by vpMbTracker::initClick(), and vpMbTracker::initFromPoints().
void vpPose::addPoints | ( | const std::vector< vpPoint > & | lP | ) |
Add (append) a list of points in the array of points.
lP | : List of points to add (append). |
Definition at line 163 of file vpPose.cpp.
|
protected |
Return 0 if success, -1 if failure.
Definition at line 297 of file vpPoseDementhon.cpp.
void vpPose::clearPoint | ( | ) |
Delete the array of point
Definition at line 133 of file vpPose.cpp.
Referenced by vpMbTracker::initClick().
|
static |
Compute the pose of a planar object from corresponding 2D-3D point coordinates and depth map. Depth map is here used to estimate the 3D plane of the object.
[in] | depthMap | : Depth map aligned to the color image from where corners are extracted. |
[in] | corners | : Vector of 2D pixel coordinates of the object in an image. |
[in] | colorIntrinsics | : Camera parameters used to convert corners from pixel to meters. |
[in] | point3d | : Vector of 3D points corresponding to the model of the planar object. |
[out] | cMo | : Computed pose. |
[out] | confidence_index | : Confidence index in range [0, 1]. When values are close to 1, it means that pose estimation confidence is high. Values close to 0 indicate that pose is not well estimated. This confidence index corresponds to the product between the normalized number of depth data covering the tag and the normalized M-estimator weights returned by the robust estimation of the tag 3D plane. |
The following code snippet implemented in tutorial-apriltag-detector-live-rgbd-realsense.cpp shows how to use this function to estimate the pose of an AprilTag using this method:
Definition at line 250 of file vpPoseRGBD.cpp.
bool vpPose::computePose | ( | vpPoseMethodType | method, |
vpHomogeneousMatrix & | cMo, | ||
bool(*)(const vpHomogeneousMatrix &) | func = NULL |
||
) |
Compute the pose according to the desired method which are:
Definition at line 373 of file vpPose.cpp.
Referenced by vpMbTracker::initClick(), and vpMbTracker::initFromPoints().
|
static |
Compute the number of RANSAC iterations to ensure with a probability p that at least one of the random samples of s points is free from outliers.
probability | : Probability that at least one of the random samples is free from outliers (typically p=0.99). |
epsilon | : Probability that a selected point is an outlier (between 0 and 1). |
sampleSize | : Minimum number of points to estimate the model (4 for a pose estimation). |
maxIterations | : Upper bound on the number of iterations or -1 for INT_MAX. |
maxIterations
if it exceeds the desired upper bound or INT_MAX if maxIterations=-1. Definition at line 591 of file vpPoseRansac.cpp.
double vpPose::computeResidual | ( | const vpHomogeneousMatrix & | cMo | ) | const |
Compute and return the sum of squared residuals expressed in meter^2 for the pose matrix cMo.
cMo | : Input pose. The matrix that defines the pose to be tested. |
Definition at line 335 of file vpPose.cpp.
Referenced by vpMbTracker::initClick(), and vpMbTracker::initFromPoints().
|
protected |
Compute and return the residual corresponding to the sum of squared residuals in meter^2 for the pose matrix cMo.
cMo | : the matrix that defines the pose to be tested. |
Definition at line 545 of file vpPoseDementhon.cpp.
bool vpPose::coplanar | ( | int & | coplanar_plane_type | ) |
Test the coplanarity of the set of points
coplanar_plane_type | 1: if plane x=cst 2: if plane y=cst 3: if plane z=cst 4: if the points are collinear. 0: any other plane |
Definition at line 183 of file vpPose.cpp.
|
static |
Display in the image I the pose represented by its homogenous transformation cMo as a 3 axis frame.
I | Image where the pose is displayed in overlay. |
cMo | Considered pose to display. |
cam | Camera parameters associated to image I. |
size | length in meter of the axis that will be displayed |
col | Color used to display the 3 axis. If vpColor::none, red, green and blue will represent x-axiw, y-axis and z-axis respectively. |
Definition at line 488 of file vpPose.cpp.
|
static |
Display in the image I the pose represented by its homogenous transformation cMo as a 3 axis frame.
I | Image where the pose is displayed in overlay. |
cMo | Considered pose to display. |
cam | Camera parameters associated to image I. |
size | length in meter of the axis that will be displayed |
col | Color used to display the 3 axis. If vpColor::none, red, green and blue will represent x-axiw, y-axis and z-axis respectively. |
Definition at line 503 of file vpPose.cpp.
void vpPose::displayModel | ( | vpImage< unsigned char > & | I, |
vpCameraParameters & | cam, | ||
vpColor | col = vpColor::none |
||
) |
Display the coordinates of the points in the image plane that are used to compute the pose in image I.
Definition at line 512 of file vpPose.cpp.
void vpPose::displayModel | ( | vpImage< vpRGBa > & | I, |
vpCameraParameters & | cam, | ||
vpColor | col = vpColor::none |
||
) |
Display the coordinates of the points in the image plane that are used to compute the pose in image I.
Definition at line 530 of file vpPose.cpp.
|
static |
Match a vector p2D of 2D point (x,y) and a vector p3D of 3D points (X,Y,Z) using the Ransac algorithm.
At least numberOfInlierToReachAConsensus of true correspondance are required to validate the pose
The inliers are given in a vector of vpPoint listInliers.
The pose is returned in cMo.
p2D | : Vector of 2d points (x and y attributes are used). |
p3D | : Vector of 3d points (oX, oY and oZ attributes are used). |
numberOfInlierToReachAConsensus | : The minimum number of inlier to have to consider a trial as correct. |
threshold | : The maximum error allowed between the 2d points and the reprojection of its associated 3d points by the current pose (in meter). |
ninliers | : Number of inliers found for the best solution. |
listInliers | : Vector of points (2d and 3d) that are inliers for the best solution. |
cMo | : The computed pose (best solution). |
maxNbTrials | : Maximum number of trials before considering a solution fitting the required numberOfInlierToReachAConsensus and threshold cannot be found. |
useParallelRansac | : If true, use parallel RANSAC version (if C++11 is available). |
nthreads | : Number of threads to use, if 0 the number of CPU threads will be determined. |
func | : Pointer to a function that takes in parameter a vpHomogeneousMatrix and returns true if the pose check is OK or false otherwise |
Definition at line 659 of file vpPoseRansac.cpp.
|
inline |
Get the covariance matrix computed in the Virtual Visual Servoing approach.
|
inline |
Get the number of threads for the parallel RANSAC implementation.
|
inline |
|
inline |
|
inline |
|
inline |
void vpPose::init | ( | void | ) |
Definition at line 62 of file vpPose.cpp.
void vpPose::poseDementhonNonPlan | ( | vpHomogeneousMatrix & | cMo | ) |
Compute the pose using Dementhon approach for non planar objects. This is a direct implementation of the algorithm proposed by Dementhon and Davis in their 1995 paper [Dementhon95].
Definition at line 107 of file vpPoseDementhon.cpp.
void vpPose::poseDementhonPlan | ( | vpHomogeneousMatrix & | cMo | ) |
Compute the pose using Dementhon approach for planar objects this is a direct implementation of the algorithm proposed by Dementhon in his PhD.
Definition at line 420 of file vpPoseDementhon.cpp.
|
static |
Carries out the camera pose the image of a rectangle and the intrinsec parameters, the length on x axis is known but the proprtion of the rectangle are unknown.
This method is taken from "Markerless Tracking using Planar Structures in the Scene" by Gilles Simon. The idea is to compute the homography H giving the image point of the rectangle by associating them with the coordinates (0,0)(1,0)(1,1/s)(0,1/s) (the rectangle is on the Z=0 plane). If K is the intrinsec parameters matrix, we have s = ||Kh1||/ ||Kh2||. s gives us the proportion of the rectangle
p1,p2,p3,p4 | the image of the corners of the rectangle (respectively the image of (0,0),(lx,0),(lx,lx/s) and (0,lx/s)) (input) |
cam | the camera used (input) |
lx | the rectangle size on the x axis (input) |
cMo | the camera pose (output) |
Definition at line 563 of file vpPose.cpp.
void vpPose::poseLagrangeNonPlan | ( | vpHomogeneousMatrix & | cMo | ) |
Definition at line 532 of file vpPoseLagrange.cpp.
References vpException::dimensionError, vpException::divideByZeroError, vpPoint::get_oX(), vpPoint::get_oY(), vpPoint::get_oZ(), vpPoint::get_x(), vpPoint::get_y(), listP, npt, and vpColVector::sumSquare().
void vpPose::poseLagrangePlan | ( | vpHomogeneousMatrix & | cMo | ) |
Compute the pose of a planar object using Lagrange approach.
cMo | : Estimated pose. No initialisation is requested to estimate cMo. |
Definition at line 255 of file vpPoseLagrange.cpp.
void vpPose::poseLowe | ( | vpHomogeneousMatrix & | cMo | ) |
Compute the pose using the Lowe non linear approach it consider the minimization of a residual using the levenberg marquartd approach.
The approach has been proposed by D.G Lowe in 1992 paper [Lowe92a].
Definition at line 261 of file vpPoseLowe.cpp.
bool vpPose::poseRansac | ( | vpHomogeneousMatrix & | cMo, |
bool(*)(const vpHomogeneousMatrix &) | func = NULL |
||
) |
Compute the pose using the Ransac approach.
cMo | : Computed pose |
func | : Pointer to a function that takes in parameter a vpHomogeneousMatrix and returns true if the pose check is OK or false otherwise |
Definition at line 319 of file vpPoseRansac.cpp.
void vpPose::poseVirtualVS | ( | vpHomogeneousMatrix & | cMo | ) |
Compute the pose using virtual visual servoing approach.
This approach is described in [Marchand02c].
Definition at line 55 of file vpPoseVirtualVisualServoing.cpp.
void vpPose::poseVirtualVSrobust | ( | vpHomogeneousMatrix & | cMo | ) |
Compute the pose using virtual visual servoing approach and a robust control law.
This approach is described in [Comport06b].
Definition at line 160 of file vpPoseVirtualVisualServoing.cpp.
void vpPose::printPoint | ( | ) |
Definition at line 467 of file vpPose.cpp.
References vpTracker::cP, listP, vpForwardProjection::oP, vpTracker::p, and vpColVector::t().
|
inline |
void vpPose::setDistanceToPlaneForCoplanarityTest | ( | double | d | ) |
Definition at line 170 of file vpPose.cpp.
|
inline |
Set the number of threads for the parallel RANSAC implementation.
|
inline |
Set RANSAC filter flag.
flag | : RANSAC flag to use to prefilter or perform degenerate configuration check. |
|
inline |
|
inline |
|
inline |
|
inline |
|
protected |
std::list<vpPoint> vpPose::listP |
Array of point (use here class vpPoint)
Definition at line 109 of file vpPose.h.
Referenced by poseLagrangeNonPlan(), and printPoint().
unsigned int vpPose::npt |
Number of point used in pose computation.
Definition at line 108 of file vpPose.h.
Referenced by poseLagrangeNonPlan().