39 #ifndef PCL_REGISTRATION_TRANSFORMATION_ESTIMATION_SVD_SCALE_HPP_
40 #define PCL_REGISTRATION_TRANSFORMATION_ESTIMATION_SVD_SCALE_HPP_
43 template <
typename Po
intSource,
typename Po
intTarget,
typename Scalar>
void
45 const Eigen::Matrix<Scalar, Eigen::Dynamic, Eigen::Dynamic> &cloud_src_demean,
46 const Eigen::Matrix<Scalar, 4, 1> ¢roid_src,
47 const Eigen::Matrix<Scalar, Eigen::Dynamic, Eigen::Dynamic> &cloud_tgt_demean,
48 const Eigen::Matrix<Scalar, 4, 1> ¢roid_tgt,
49 Matrix4 &transformation_matrix)
const
51 transformation_matrix.setIdentity ();
54 Eigen::Matrix<Scalar, 3, 3> H = (cloud_src_demean * cloud_tgt_demean.transpose ()).topLeftCorner (3, 3);
57 Eigen::JacobiSVD<Eigen::Matrix<Scalar, 3, 3> > svd (H, Eigen::ComputeFullU | Eigen::ComputeFullV);
58 Eigen::Matrix<Scalar, 3, 3> u = svd.matrixU ();
59 Eigen::Matrix<Scalar, 3, 3> v = svd.matrixV ();
62 if (u.determinant () * v.determinant () < 0)
64 for (
int x = 0; x < 3; ++x)
68 Eigen::Matrix<Scalar, 3, 3> R = v * u.transpose ();
71 Eigen::Matrix<Scalar, 4, 4> R4;
72 R4.block (0, 0, 3, 3) = R;
78 Eigen::Matrix<Scalar, Eigen::Dynamic, Eigen::Dynamic> src_ = R4 * cloud_src_demean;
81 double sum_ss = 0.0f, sum_tt = 0.0f, sum_tt_ = 0.0f;
82 for (
unsigned corrIdx = 0; corrIdx < cloud_src_demean.cols (); ++corrIdx)
84 sum_ss += cloud_src_demean (0, corrIdx) * cloud_src_demean (0, corrIdx);
85 sum_ss += cloud_src_demean (1, corrIdx) * cloud_src_demean (1, corrIdx);
86 sum_ss += cloud_src_demean (2, corrIdx) * cloud_src_demean (2, corrIdx);
88 sum_tt += cloud_tgt_demean (0, corrIdx) * cloud_tgt_demean (0, corrIdx);
89 sum_tt += cloud_tgt_demean (1, corrIdx) * cloud_tgt_demean (1, corrIdx);
90 sum_tt += cloud_tgt_demean (2, corrIdx) * cloud_tgt_demean (2, corrIdx);
92 sum_tt_ += cloud_tgt_demean (0, corrIdx) * src_ (0, corrIdx);
93 sum_tt_ += cloud_tgt_demean (1, corrIdx) * src_ (1, corrIdx);
94 sum_tt_ += cloud_tgt_demean (2, corrIdx) * src_ (2, corrIdx);
97 scale1 = sqrt (sum_tt / sum_ss);
98 scale2 = sum_tt_ / sum_ss;
100 transformation_matrix.topLeftCorner (3, 3) = scale * R;
101 const Eigen::Matrix<Scalar, 3, 1> Rc (scale * R * centroid_src.head (3));
102 transformation_matrix.block (0, 3, 3, 1) = centroid_tgt. head (3) - Rc;