/*! Initialize a point feature with polar coordinates \f$(\rho,\theta)\f$ using the coordinates of the point \f$(x,y,Z)\f$, where \f$(x,y)\f$ correspond to the perspective projection of the point in the image plane and \f$Z\f$ the 3D depth of the point in the camera frame. The values of \f$(x,y,Z)\f$ are expressed in meters. This function intends to introduce noise in the conversion from cartesian to polar coordinates. Cartesian \f$(x,y)\f$ coordinates are first converted in pixel coordinates in the image using \e goodCam camera parameters. Then, the pixels coordinates of the point are converted back to cartesian coordinates \f$(x^{'},y^{'})\f$ using the noisy camera parameters \e wrongCam. From these new coordinates in the image plane, the polar coordinates are computed by: \f[\rho = \sqrt{x^2+y^2} \hbox{,}\; \; \theta = \arctan \frac{y}{x}\f] \param s : Visual feature \f$(\rho,\theta)\f$ and \f$Z\f$ to initialize. \param goodCam : Camera parameters used to introduce noise. These parameters are used to convert cartesian coordinates of the point \e p in the image plane in pixel coordinates. \param wrongCam : Camera parameters used to introduce noise. These parameters are used to convert pixel coordinates of the point in cartesian coordinates of the point in the image plane. \param p : A point with \f$(x,y)\f$ cartesian coordinates in the image plane corresponding to the camera perspective projection, and with 3D depth \f$Z\f$. */ void vpFeatureBuilder::create(vpFeaturePointPolar &s, const vpCameraParameters &goodCam, const vpCameraParameters &wrongCam, const vpPoint &p) { try { double x = p.get_x(); double y = p.get_y(); s.set_Z( p.get_Z() ); double u=0, v=0; vpMeterPixelConversion::convertPoint(goodCam, x, y, u, v); vpPixelMeterConversion::convertPoint(wrongCam, u, v, x, y); double rho = sqrt(x*x + y*y); double theta = atan2(y, x); s.set_rho(rho) ; s.set_theta(theta) ; } catch(...) { vpERROR_TRACE("Error caught") ; throw ; } }
/*! Create a vpColVector of a projected point. \param p : Point to project. \param v : Resulting vector. \param K : Camera parameters. */ void vpMbScanLine::createVectorFromPoint(const vpPoint &p, vpColVector &v, const vpCameraParameters &K) { v = vpColVector(3); v[0] = p.get_X() * K.get_px() + K.get_u0() * p.get_Z(); v[1] = p.get_Y() * K.get_py() + K.get_v0() * p.get_Z(); v[2] = p.get_Z(); }
/*! Interpolate two vpPoints. \param a : first point. \param b : second point. \param alpha : interpolation factor. \return Interpolated vpPoint. */ vpPoint vpMbScanLine::mix(const vpPoint &a, const vpPoint &b, double alpha) { vpPoint res; res.set_X(a.get_X() + ( b.get_X() - a.get_X() ) * alpha); res.set_Y(a.get_Y() + ( b.get_Y() - a.get_Y() ) * alpha); res.set_Z(a.get_Z() + ( b.get_Z() - a.get_Z() ) * alpha); return res; }
/*! Initialize a vpFeatureVanishingPoint thanks to a vpPoint. The vpFeatureVanishingPoint is initialized thanks to the parameters of the point in the image plan. All the parameters are given in meter. \param s : Visual feature to initialize. \param t : The vpPoint used to create the vpFeatureVanishingPoint. */ void vpFeatureBuilder::create(vpFeatureVanishingPoint &s, const vpPoint &t) { try { s.set_x( t.get_x()) ; s.set_y( t.get_y()) ; } catch(...) { vpERROR_TRACE("Cannot create vanishing point feature") ; throw ; } }
/*! Build a vpMbtDistanceSphere thanks to its center, 3 points (including the center) with coordinates expressed in the object frame and defining the plane that contain the circle and its radius. \param _p1 : Center of the circle. \param _p2,_p3 : Two points on the plane containing the circle. With the center of the circle we have 3 points defining the plane that contains the circle. \param r : Radius of the circle. */ void vpMbtDistanceSphere::buildFrom(const vpPoint &_p1, const double r) { sphere = new vpSphere ; p1 = new vpPoint ; // Get the points *p1 = _p1; // Get the radius radius = r; // Build our sphere sphere->setWorldCoordinates(_p1.get_oX(), _p1.get_oY(), _p1.get_oZ(), r); }
/*! From the coordinates of the point in image plane b and the homography between image a and b computes the coordinates of the point in image plane a. \param b_P : 2D coordinates of the point in the image plane b. \return A point with 2D coordinates in the image plane a. */ vpPoint vpHomography::operator*(const vpPoint& b_P) const { vpPoint a_P ; vpColVector v(3),v1(3) ; v[0] = b_P.get_x() ; v[1] = b_P.get_y() ; v[2] = b_P.get_w() ; v1[0] = (*this)[0][0]*v[0] + (*this)[0][1]*v[1]+ (*this)[0][2]*v[2] ; v1[1] = (*this)[1][0]*v[0] + (*this)[1][1]*v[1]+ (*this)[1][2]*v[2] ; v1[2] = (*this)[2][0]*v[0] + (*this)[2][1]*v[1]+ (*this)[2][2]*v[2] ; // v1 = M*v ; a_P.set_x(v1[0]) ; a_P.set_y(v1[1]) ; a_P.set_w(v1[2]) ; return a_P ; }
/*! Create a vpMbScanLineEdge from two points while ordering them. \param a : First point of the line. \param b : Second point of the line. \return Resulting vpMbScanLineEdge. */ vpMbScanLine::vpMbScanLineEdge vpMbScanLine::makeMbScanLineEdge(const vpPoint &a, const vpPoint &b) { vpColVector _a(3); vpColVector _b(3); _a[0] = std::ceil((a.get_X() * 1e8) * 1e-6); _a[1] = std::ceil((a.get_Y() * 1e8) * 1e-6); _a[2] = std::ceil((a.get_Z() * 1e8) * 1e-6); _b[0] = std::ceil((b.get_X() * 1e8) * 1e-6); _b[1] = std::ceil((b.get_Y() * 1e8) * 1e-6); _b[2] = std::ceil((b.get_Z() * 1e8) * 1e-6); bool b_comp = false; for(unsigned int i = 0 ; i < 3 ; ++i) if (_a[i] < _b[i]) { b_comp = true; break; } else if(_a[i] > _b[i]) break; if (b_comp) return std::make_pair(_a, _b); return std::make_pair(_b, _a); }
/*! Initialize a point feature with polar coordinates \f$(\rho,\theta)\f$ using the coordinates of the point \f$(x,y,Z)\f$, where \f$(x,y)\f$ correspond to the perspective projection of the point in the image plane and \f$Z\f$ the 3D depth of the point in the camera frame. The values of \f$(x,y,Z)\f$ are expressed in meters. From the coordinates in the image plane, the polar coordinates are computed by: \f[\rho = \sqrt{x^2+y^2} \hbox{,}\; \; \theta = \arctan \frac{y}{x}\f] \param s : Visual feature \f$(\rho,\theta)\f$ and \f$Z\f$ to initialize. \param p : A point with \f$(x,y)\f$ cartesian coordinates in the image plane corresponding to the camera perspective projection, and with 3D depth \f$Z\f$. */ void vpFeatureBuilder::create(vpFeaturePointPolar &s, const vpPoint &p) { try { double x = p.get_x(); double y = p.get_y(); double rho = sqrt(x*x + y*y); double theta = atan2(y, x); s.set_rho(rho) ; s.set_theta(theta) ; s.set_Z( p.get_Z() ) ; if (s.get_Z() < 0) { vpERROR_TRACE("Point is behind the camera ") ; std::cout <<"Z = " << s.get_Z() << std::endl ; throw(vpFeatureException(vpFeatureException::badInitializationError, "Point is behind the camera ")) ; } if (fabs(s.get_Z()) < 1e-6) { vpERROR_TRACE("Point Z coordinates is null ") ; std::cout <<"Z = " << s.get_Z() << std::endl ; throw(vpFeatureException(vpFeatureException::badInitializationError, "Point Z coordinates is null")) ; } } catch(...) { vpERROR_TRACE("Error caught") ; throw ; } }
void kltFbTracker::genFeaturesForTrackOnFace(std::vector<vpPoint>& features, int faceID, int numOfPtsPerFace) { const vpPoint& vp1 = pyg[faceID].p[0], vp2 = pyg[faceID].p[1], vp3 = pyg[faceID].p[2], vp4 = pyg[faceID].p[3]; float step = 1.0 / numOfPtsPerFace; // avoid the edge for (int i = 1; i < numOfPtsPerFace; i++) for (int j = 1; j < numOfPtsPerFace; j++) { vpPoint fp; float alpha = i * step; float beta = j * step; fp.set_oX((alpha + beta - 1) * vp2.get_oX() + (1 - alpha) * vp1.get_oX() + (1 - beta) * vp3.get_oX()); fp.set_oY((alpha + beta - 1) * vp2.get_oY() + (1 - alpha) * vp1.get_oY() + (1 - beta) * vp3.get_oY()); fp.set_oZ((alpha + beta - 1) * vp2.get_oZ() + (1 - alpha) * vp1.get_oZ() + (1 - beta) * vp3.get_oZ()); features.push_back(fp); } }
/*! Unary function to convert the 3D coordinates in the object frame to a cv::Point3d. \param point : Point to convert. \return A cv::Point3d with the 3D coordinates stored in vpPoint in the object frame. */ cv::Point3d vpConvert::vpObjectPointToPoint3d(const vpPoint &point) { return cv::Point3d(point.get_oX(), point.get_oY(), point.get_oZ()); }
/*! Unary function to convert the 3D coordinates in the object frame to a cv::Point3f. \param point : Point to convert. \return A cv::Point3f with the 3D coordinates stored in vpPoint in the object frame. */ cv::Point3f vpConvert::vpObjectPointToPoint3f(const vpPoint &point) { return cv::Point3f((float) point.get_oX(), (float) point.get_oY(), (float) point.get_oZ()); }
/*! Unary function to convert the 3D coordinates in the camera frame to a cv::Point3d. \param point : Point to convert. \return A cv::Point3d with the 3D coordinates stored in vpPoint in the camera frame. */ cv::Point3d vpConvert::vpCamPointToPoint3d(const vpPoint &point) { return cv::Point3d(point.get_X(), point.get_Y(), point.get_Z()); }
/*! Unary function to convert the 3D coordinates in the camera frame to a cv::Point3f. \param point : Point to convert. \return A cv::Point3f with the 3D coordinates stored in vpPoint in the camera frame. */ cv::Point3f vpConvert::vpCamPointToPoint3f(const vpPoint &point) { return cv::Point3f((float) point.get_X(), (float) point.get_Y(), (float) point.get_Z()); }
void vpMbtDistanceKltCylinder::buildFrom(const vpPoint &p1, const vpPoint &p2, const double &r) { p1Ext = p1; p2Ext = p2; vpColVector ABC(3); vpColVector V1(3); vpColVector V2(3); V1[0] = p1.get_oX(); V1[1] = p1.get_oY(); V1[2] = p1.get_oZ(); V2[0] = p2.get_oX(); V2[1] = p2.get_oY(); V2[2] = p2.get_oZ(); // Get the axis of the cylinder ABC = V1-V2; // Build our extremity circles circle1.setWorldCoordinates(ABC[0],ABC[1],ABC[2],p1.get_oX(),p1.get_oY(),p1.get_oZ(),r); circle2.setWorldCoordinates(ABC[0],ABC[1],ABC[2],p2.get_oX(),p2.get_oY(),p2.get_oZ(),r); // Build our cylinder cylinder.setWorldCoordinates(ABC[0],ABC[1],ABC[2],(p1.get_oX()+p2.get_oX())/2.0,(p1.get_oY()+p2.get_oY())/2.0,(p1.get_oZ()+p2.get_oZ())/2.0,r); }
/*! Build a 3D plane thanks to 3 points and stores it in \f$ plane \f$. \param P : The first point to define the plane \param Q : The second point to define the plane \param R : The third point to define the plane \param plane : The vpPlane instance used to store the computed plane equation. */ void buildPlane(vpPoint &P, vpPoint &Q, vpPoint &R, vpPlane &plane) { vpColVector a(3); vpColVector b(3); vpColVector n(3); //Calculate vector corresponding to PQ a[0]=P.get_oX()-Q.get_oX(); a[1]=P.get_oY()-Q.get_oY(); a[2]=P.get_oZ()-Q.get_oZ(); //Calculate vector corresponding to PR b[0]=P.get_oX()-R.get_oX(); b[1]=P.get_oY()-R.get_oY(); b[2]=P.get_oZ()-R.get_oZ(); //Calculate normal vector to plane PQ x PR n=vpColVector::cross(a,b); //Equation of the plane is given by: double A = n[0]; double B = n[1]; double C = n[2]; double D=-(A*P.get_oX()+B*P.get_oY()+C*P.get_oZ()); double norm = sqrt(A*A+B*B+C*C) ; plane.setA(A/norm) ; plane.setB(B/norm) ; plane.setC(C/norm) ; plane.setD(D/norm) ; }
/*! Compute the norm of two vpPoints. \param a : first point. \param b : second point. \return Resulting norm. */ double vpMbScanLine::norm(const vpPoint &a, const vpPoint &b) { return sqrt(vpMath::sqr(a.get_X()-b.get_X()) + vpMath::sqr(a.get_Y() - b.get_Y()) + vpMath::sqr(a.get_Z() - b.get_Z())); }
/*! Get the clipped points according to a plane equation. \param cam : camera parameters \param p1 : First extremity of the line. \param p2 : Second extremity of the line. \param p1Clipped : Resulting p1. \param p2Clipped : Resulting p2. \param p1ClippedInfo : Resulting clipping flag for p1. \param p2ClippedInfo : Resulting clipping flag for p2. \param A : Param A from plane equation. \param B : Param B from plane equation. \param C : Param C from plane equation. \param D : Param D from plane equation. \param flag : flag specifying the clipping used when calling this function. \return True if the points have been clipped, False otherwise */ bool vpPolygon3D::getClippedPointsFovGeneric(const vpPoint &p1, const vpPoint &p2, vpPoint &p1Clipped, vpPoint &p2Clipped, unsigned int &p1ClippedInfo, unsigned int &p2ClippedInfo, const vpColVector &normal, const unsigned int &flag) { vpRowVector p1Vec(3); p1Vec[0] = p1.get_X(); p1Vec[1] = p1.get_Y(); p1Vec[2] = p1.get_Z(); p1Vec = p1Vec.normalize(); vpRowVector p2Vec(3); p2Vec[0] = p2.get_X(); p2Vec[1] = p2.get_Y(); p2Vec[2] = p2.get_Z(); p2Vec = p2Vec.normalize(); if((clippingFlag & flag) == flag){ double beta1 = acos( p1Vec * normal ); double beta2 = acos( p2Vec * normal ); // std::cout << beta1 << " && " << beta2 << std::endl; // if(!(beta1 < M_PI / 2.0 && beta2 < M_PI / 2.0)) if(beta1 < M_PI / 2.0 && beta2 < M_PI / 2.0) return false; else if (beta1 < M_PI / 2.0 || beta2 < M_PI / 2.0){ vpPoint pClipped; double t = -(normal[0] * p1.get_X() + normal[1] * p1.get_Y() + normal[2] * p1.get_Z()); t = t / ( normal[0] * (p2.get_X() - p1.get_X()) + normal[1] * (p2.get_Y() - p1.get_Y()) + normal[2] * (p2.get_Z() - p1.get_Z()) ); pClipped.set_X((p2.get_X() - p1.get_X())*t + p1.get_X()); pClipped.set_Y((p2.get_Y() - p1.get_Y())*t + p1.get_Y()); pClipped.set_Z((p2.get_Z() - p1.get_Z())*t + p1.get_Z()); if(beta1 < M_PI / 2.0){ p1ClippedInfo = p1ClippedInfo | flag; p1Clipped = pClipped; } else{ p2ClippedInfo = p2ClippedInfo | flag; p2Clipped = pClipped; } } } return true; }
bool vpPolygon3D::getClippedPointsDistance(const vpPoint &p1, const vpPoint &p2, vpPoint &p1Clipped, vpPoint &p2Clipped, unsigned int &p1ClippedInfo, unsigned int &p2ClippedInfo, const unsigned int &flag, const double &distance) { // Since p1 and p1Clipped can be the same object as well as p2 and p2Clipped // to avoid a valgrind "Source and destination overlap in memcpy" error, // we introduce a two temporary points. vpPoint p1Clipped_, p2Clipped_; p1Clipped_ = p1; p2Clipped_ = p2; p1Clipped = p1Clipped_; p2Clipped = p2Clipped_; bool test1 = (p1Clipped.get_Z() < distance && p2Clipped.get_Z() < distance); if(flag == vpPolygon3D::FAR_CLIPPING) test1 = (p1Clipped.get_Z() > distance && p2Clipped.get_Z() > distance); bool test2 = (p1Clipped.get_Z() < distance || p2Clipped.get_Z() < distance); if(flag == vpPolygon3D::FAR_CLIPPING) test2 = (p1Clipped.get_Z() > distance || p2Clipped.get_Z() > distance); bool test3 = (p1Clipped.get_Z() < distance); if(flag == vpPolygon3D::FAR_CLIPPING) test3 = (p1Clipped.get_Z() > distance); if(test1) return false; else if(test2){ vpPoint pClippedNear; double t; t = (p2Clipped.get_Z() - p1Clipped.get_Z()); t = (distance - p1Clipped.get_Z()) / t; pClippedNear.set_X((p2Clipped.get_X() - p1Clipped.get_X())*t + p1Clipped.get_X()); pClippedNear.set_Y((p2Clipped.get_Y() - p1Clipped.get_Y())*t + p1Clipped.get_Y()); pClippedNear.set_Z(distance); if(test3){ p1Clipped = pClippedNear; if(flag == vpPolygon3D::FAR_CLIPPING) p1ClippedInfo = p1ClippedInfo | vpPolygon3D::FAR_CLIPPING; else p1ClippedInfo = p1ClippedInfo | vpPolygon3D::NEAR_CLIPPING; } else{ p2Clipped = pClippedNear; if(flag == vpPolygon3D::FAR_CLIPPING) p2ClippedInfo = p2ClippedInfo | vpPolygon3D::FAR_CLIPPING; else p2ClippedInfo = p2ClippedInfo | vpPolygon3D::NEAR_CLIPPING; } } return true; }
/*! \brief 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 \param 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) \param cam: the camera used (input) \param lx: the rectangle size on the x axis (input) \param cMo: the camera pose (output) \return int : OK if no pb occurs */ double vpPose::poseFromRectangle(vpPoint &p1,vpPoint &p2, vpPoint &p3,vpPoint &p4, double lx, vpCameraParameters & cam, vpHomogeneousMatrix & cMo) { double rectx[4] ; double recty[4] ; rectx[0]= 0 ; recty[0]=0 ; rectx[1]=1 ; recty[1]=0 ; rectx[2]=1 ; recty[2]=1 ; rectx[3]=0 ; recty[3]=1 ; double irectx[4] ; double irecty[4] ; irectx[0]=(p1.get_x()) ; irecty[0]=(p1.get_y()) ; irectx[1]=(p2.get_x()) ; irecty[1]=(p2.get_y()) ; irectx[2]=(p3.get_x()) ; irecty[2]=(p3.get_y()) ; irectx[3]=(p4.get_x()) ; irecty[3]=(p4.get_y()) ; //calcul de l'homographie vpMatrix H(3,3); vpHomography hom; // vpHomography::HartleyDLT(4,rectx,recty,irectx,irecty,hom); vpHomography::HLM(4,rectx,recty,irectx,irecty,1,hom); for (unsigned int i=0 ; i < 3 ; i++) for(unsigned int j=0 ; j < 3 ; j++) H[i][j] = hom[i][j] ; //calcul de s = ||Kh1||/ ||Kh2|| =ratio (length on x axis/ length on y axis) vpColVector kh1(3); vpColVector kh2(3); vpMatrix K(3,3); K = cam.get_K(); K.setIdentity(); vpMatrix Kinv =K.pseudoInverse(); vpMatrix KinvH =Kinv*H; kh1=KinvH.column(1); kh2=KinvH.column(2); double s= sqrt(kh1.sumSquare())/sqrt(kh2.sumSquare()); vpMatrix D(3,3); D.setIdentity(); D[1][1]=1/s; vpMatrix cHo=H*D; //Calcul de la rotation et de la translation // PoseFromRectangle(p1,p2,p3,p4,1/s,lx,cam,cMo ); p1.setWorldCoordinates(0,0,0) ; p2.setWorldCoordinates(lx,0,0) ; p3.setWorldCoordinates(lx,lx/s,0) ; p4.setWorldCoordinates(0,lx/s,0) ; vpPose P ; P.addPoint(p1) ; P.addPoint(p2) ; P.addPoint(p3) ; P.addPoint(p4) ; P.computePose(vpPose::DEMENTHON_LOWE,cMo) ; return lx/s ; }
/*! Test the visibility of a line. As a result, a subsampled line of the given one with all its visible parts. \param a : First point of the line. \param b : Second point of the line. \param lines : List of lines corresponding of the visible parts of the given line. \param displayResults : True if the results have to be displayed. False otherwise. */ void vpMbScanLine::queryLineVisibility(vpPoint a, vpPoint b, std::vector<std::pair<vpPoint, vpPoint> > &lines, const bool &displayResults) { vpColVector _a, _b; createVectorFromPoint(a, _a, K); createVectorFromPoint(b, _b, K); double x0 = _a[0] / _a[2]; double y0 = _a[1] / _a[2]; double z0 = _a[2]; double x1 = _b[0] / _b[2]; double y1 = _b[1] / _b[2]; double z1 = _b[2]; vpMbScanLineEdge edge = makeMbScanLineEdge(a, b); lines.clear(); if(displayResults){ #if defined(VISP_HAVE_X11) && defined(DEBUG_DISP) double i1(0.0), j1(0.0), i2(0.0), j2(0.0); a.project(); b.project(); vpMeterPixelConversion::convertPoint(K,a.get_x(), a.get_y(),j1,i1); vpMeterPixelConversion::convertPoint(K,b.get_x(), b.get_y(),j2,i2); vpDisplay::displayLine(linedebugImg,i1,j1,i2,j2,vpColor::yellow,3); #endif } if (!visibility_samples.count(edge)) return; // Initialized as the biggest difference between the two points is on the X-axis double *v0(&x0), *w0(&z0); double *v1(&x1), *w1(&z1); unsigned int size(w); if (std::fabs(y0 - y1) > std::fabs(x0 - x1)) // Test if the biggest difference is on the Y-axis { v0 = &y0; v1 = &y1; size = h; } if (*v0 > *v1) { std::swap(v0, v1); std::swap(w0, w1); std::swap(a, b); } //if (*v0 >= size - 1 || *v1 < 0 || *v1 == *v0) if (*v0 >= size - 1 || *v1 < 0 || std::fabs(*v1 - *v0) <= std::numeric_limits<double>::epsilon()) return; const int _v0 = std::max(0, int(std::ceil(*v0))); const int _v1 = std::min<int>((int)(size - 1), (int)(std::ceil(*v1) - 1)); const std::set<int> &visible_samples = visibility_samples[edge]; int last = _v0; vpPoint line_start; vpPoint line_end; bool b_line_started = false; for(std::set<int>::const_iterator it = visible_samples.begin() ; it != visible_samples.end() ; ++it) { const int v = *it; const double alpha = getAlpha(v, (*v0) * (*w0), (*w0), (*v1) * (*w1), (*w1)); const vpPoint p = mix(a, b, alpha); if (last + 1 != v) { if(b_line_started) lines.push_back(std::make_pair(line_start, line_end)); b_line_started = false; } if (v == _v0) { line_start = a; line_end = p; b_line_started = true; } else if (v == _v1) { line_end = b; if (!b_line_started) line_start = p; b_line_started = true; } else { line_end = p; if (!b_line_started) line_start = p; b_line_started = true; } last = v; } if (b_line_started) lines.push_back(std::make_pair(line_start, line_end)); if(displayResults){ #if defined(VISP_HAVE_X11) && defined(DEBUG_DISP) double i1(0.0), j1(0.0), i2(0.0), j2(0.0); for(unsigned int i = 0 ; i < lines.size() ; i++){ lines[i].first.project(); lines[i].second.project(); vpMeterPixelConversion::convertPoint(K,lines[i].first.get_x(), lines[i].first.get_y(),j1,i1); vpMeterPixelConversion::convertPoint(K,lines[i].second.get_x(), lines[i].second.get_y(),j2,i2); vpDisplay::displayLine(linedebugImg,i1,j1,i2,j2,vpColor::red,3); } vpDisplay::flush(linedebugImg); #endif } }