【原文:http://www.cnblogs.com/yhlx125/p/5709157.html】
泛化的ICP算法,通过协方差矩阵起到类似于权重的作用,消除某些不好的对应点在求解过程中的作用。不过可以囊括Point to Point,Point to Plane的ICP算法,真正的是泛化的ICP。牛!
近似特征值计算
这个的原理被想复杂了,就是特征值分解的逆步骤,形成了三个正交的向量,epsilon是最小的特征值,法向量是最小的特征向量。
本来求法向量的过程就是根据近邻的k个点,利用主成分分析PCA进行计算得到特征值最小的那个特征向量作为法向量。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
|
/** \brief Compute GICP-style covariance matrices given a point cloud and
* the corresponding surface normals.
* \param[in] cloud Point cloud containing the XYZ coordinates,
* \param[in] normals Point cloud containing the corresponding surface normals.
* \param[out] covariances Vector of computed covariances.
* \param[in] Optional: Epsilon for the expected noise along the surface normal (default: 0.001)
*/
template
<
typename
PointT,
typename
PointNT>
inline
void
computeApproximateCovariances(
const
pcl::PointCloud<PointT>& cloud,
const
pcl::PointCloud<PointNT>& normals,
std::vector<Eigen::Matrix3d, Eigen::aligned_allocator<Eigen::Matrix3d> >& covariances,
double
epsilon = 0.001)
{
assert
(cloud.points.size() == normals.points.size());
int
nr_points =
static_cast
<
int
>(cloud.points.size());
covariances.resize(nr_points);
for
(
int
i = 0; i < nr_points; ++i)
{
Eigen::Vector3d normal(normals.points[i].normal_x,
normals.points[i].normal_y,
normals.points[i].normal_z);
// compute rotation matrix
Eigen::Matrix3d rot;
Eigen::Vector3d y;
y << 0, 1, 0;
rot.row(2) = normal;
y = y - normal(1) * normal;
y.normalize();
rot.row(1) = y;
rot.row(0) = normal.cross(rot.row(1));
// comnpute approximate covariance
Eigen::Matrix3d cov;
cov << 1, 0, 0,
0, 1, 0,
0, 0, epsilon;
covariances[i] = rot.transpose()*cov*rot;
}
}
}
|