void debugDerivatives(const double *computed_grad, const K_OPTIM::ErrorFunction & function, const K_OPTIM::OptimizationParameter & pos)
{
TooN::Vector<> tmppos(function.numParameters());
tmppos = TooN::Zeros;
TooN::Matrix<TooN::Dynamic, 2> mat = TooN::numerical_gradient_with_errors(ComposeWrapperErrorFunction(function, pos), tmppos);
for (int i = 0; i < mat.num_rows(); ++i) {
fprintf(stderr, "%i: %e (%f) <=> %e%s\n", i, mat(i, 0), mat(i, 1), computed_grad[i], (fabs(mat(i, 0) - computed_grad[i]) > 1e-4) ? " !!!" : "");
}
}
// computes the orientation from (e1,e2,e3) -> (a,(a^b)^a,a^b), which means that b the second vector is in the a, b plane
static inline TooN::SO3<> canonicalOrientation( const TooN::Vector<3> & a, const TooN::Vector<3> & b ){
TooN::Vector<3> n = a ^ b;
if(norm_sq(n) < 1e-30)
return TooN::SO3<>();
TooN::Matrix<3> result;
result.T()[0] = unit(a);
result.T()[2] = unit(n);
result.T()[1] = result.T()[2] ^ result.T()[0];
return TooN::SO3<> (result);
}
void ofxCalibImage::guessInitialPose(ofxATANCamera &Camera){
// First, find a homography which maps the grid to the unprojected image coords
// Use the standard null-space-of-SVD-thing to find 9 homography parms
// (c.f. appendix of thesis)
int nPoints = static_cast<int>(mvGridCorners.size());
TooN::Matrix<> m2Nx9(2*nPoints, 9);
for(int n=0; n<nPoints; n++){
// First, un-project the points to the image plane
TooN::Vector<2> v2UnProj = Camera.UnProject(mvGridCorners[n].Params.v2Pos);
double u = v2UnProj[0];
double v = v2UnProj[1];
// Then fill in the matrix..
double x = mvGridCorners[n].irGridPos.x;
double y = mvGridCorners[n].irGridPos.y;
m2Nx9[n*2+0][0] = x;
m2Nx9[n*2+0][1] = y;
m2Nx9[n*2+0][2] = 1;
m2Nx9[n*2+0][3] = 0;
m2Nx9[n*2+0][4] = 0;
m2Nx9[n*2+0][5] = 0;
m2Nx9[n*2+0][6] = -x*u;
m2Nx9[n*2+0][7] = -y*u;
m2Nx9[n*2+0][8] = -u;
m2Nx9[n*2+1][0] = 0;
m2Nx9[n*2+1][1] = 0;
m2Nx9[n*2+1][2] = 0;
m2Nx9[n*2+1][3] = x;
m2Nx9[n*2+1][4] = y;
m2Nx9[n*2+1][5] = 1;
m2Nx9[n*2+1][6] = -x*v;
m2Nx9[n*2+1][7] = -y*v;
m2Nx9[n*2+1][8] = -v;
}
// The right null-space (should only be one) of the matrix gives the homography...
TooN::SVD<> svdHomography(m2Nx9);
TooN::Vector<9> vH = svdHomography.get_VT()[8];
TooN::Matrix<3> m3Homography;
m3Homography[0] = vH.slice<0,3>();
m3Homography[1] = vH.slice<3,3>();
m3Homography[2] = vH.slice<6,3>();
// Fix up possibly poorly conditioned bits of the homography
{
TooN::SVD<2> svdTopLeftBit(m3Homography.slice<0,0,2,2>());
TooN::Vector<2> v2Diagonal = svdTopLeftBit.get_diagonal();
m3Homography = m3Homography / v2Diagonal[0];
v2Diagonal = v2Diagonal / v2Diagonal[0];
double dLambda2 = v2Diagonal[1];
TooN::Vector<2> v2b; // This is one hypothesis for v2b ; the other is the negative.
v2b[0] = 0.0;
v2b[1] = sqrt( 1.0 - (dLambda2 * dLambda2));
TooN::Vector<2> v2aprime = v2b * svdTopLeftBit.get_VT();
TooN::Vector<2> v2a = m3Homography[2].slice<0,2>();
double dDotProd = v2a * v2aprime;
if(dDotProd>0)
m3Homography[2].slice<0,2>() = v2aprime;
else
m3Homography[2].slice<0,2>() = -v2aprime;
}
// OK, now turn homography into something 3D ...simple gram-schmidt ortho-norm
// Take 3x3 matrix H with column: abt
// And add a new 3rd column: abct
TooN::Matrix<3> mRotation;
TooN::Vector<3> vTranslation;
double dMag1 = sqrt(m3Homography.T()[0] * m3Homography.T()[0]);
m3Homography = m3Homography / dMag1;
mRotation.T()[0] = m3Homography.T()[0];
// ( all components of the first vector are removed from the second...
mRotation.T()[1] = m3Homography.T()[1] - m3Homography.T()[0]*(m3Homography.T()[0]*m3Homography.T()[1]);
mRotation.T()[1] /= sqrt(mRotation.T()[1] * mRotation.T()[1]);
mRotation.T()[2] = mRotation.T()[0]^mRotation.T()[1];
vTranslation = m3Homography.T()[2];
// Store result
mse3CamFromWorld.get_rotation()=mRotation;
mse3CamFromWorld.get_translation() = vTranslation;
};
// Publish the current tracker pose
void SystemFrontendBase::PublishPose()
{
ROS_ASSERT(mpTracker);
if (mpTracker->GetTrackingQuality() == Tracker::NONE || mpTracker->IsLost())
return;
geometry_msgs::PoseWithCovarianceStamped pose_cov_msg;
pose_cov_msg.header.stamp = mpTracker->GetCurrentTimestamp();
pose_cov_msg.header.frame_id = "vision_world";
TooN::SE3<> pose = mpTracker->GetCurrentPose().inverse();
TooN::Matrix<3> rot = pose.get_rotation().get_matrix();
TooN::Matrix<6> cov = mpTracker->GetCurrentCovariance();
// clear cross correlation
cov.slice<0, 3, 3, 3>() = TooN::Zeros;
cov.slice<3, 0, 3, 3>() = TooN::Zeros;
// Change covariance matrix frame from camera to world
cov.slice<0, 0, 3, 3>() = rot * cov.slice<0, 0, 3, 3>() * rot.T();
cov.slice<3, 3, 3, 3>() = rot * cov.slice<3, 3, 3, 3>() * rot.T();
// Some hacky heuristics here
if (mpTracker->GetTrackingQuality() == Tracker::GOOD)
{
cov = cov * 1e2;
}
else if (mpTracker->GetTrackingQuality() == Tracker::DODGY)
{
cov = cov * 1e5;
}
else if (mpTracker->GetTrackingQuality() == Tracker::BAD)
{
cov = cov * 1e8;
}
pose_cov_msg.pose.pose = util::SE3ToPoseMsg(pose);
int numElems = 6;
for (int i = 0; i < numElems; ++i)
{
for (int j = 0; j < numElems; ++j)
{
pose_cov_msg.pose.covariance[i * numElems + j] = cov(i, j);
}
}
mPoseWithCovPub.publish(pose_cov_msg);
static tf::TransformBroadcaster br;
tf::Transform transform;
tf::poseMsgToTF(pose_cov_msg.pose.pose, transform);
br.sendTransform(tf::StampedTransform(transform, mpTracker->GetCurrentTimestamp(), "vision_world", "vision_pose"));
SE3Map mPoses = mpTracker->GetCurrentCameraPoses();
geometry_msgs::PoseArray pose_array_msg;
pose_array_msg.header.stamp = pose_cov_msg.header.stamp;
pose_array_msg.header.frame_id = pose_cov_msg.header.frame_id;
pose_array_msg.poses.resize(1 + mPoses.size());
pose_array_msg.poses[0] = pose_cov_msg.pose.pose;
int i = 1;
for (SE3Map::iterator it = mPoses.begin(); it != mPoses.end(); ++it, ++i)
{
pose_array_msg.poses[i] = util::SE3ToPoseMsg(it->second.inverse());
tf::poseMsgToTF(pose_array_msg.poses[i], transform);
br.sendTransform(tf::StampedTransform(transform, mpTracker->GetCurrentTimestamp(), "vision_world", it->first));
}
mPoseArrayPub.publish(pose_array_msg);
}
