Hello,

I have a set of 2D points, let us say 10 2D points. Then I save this points in a matrix, like:

Code: Select all

Eigen::Matrix2Xf matrix;
matrix.resize(2, 10);

for(size_t i = 0; i<10; ++i){
     matrix.col(i) = Eigen::Vector2f(point(i).x, point(i).y);
}

Then I followed the instructions from ggael. (viewtopic.php?f=9&t=110265#p260370)

Code: Select all

Eigen::Matrix2Xf centered    = matrix.rowwise() - matrix.colwise().mean();
Eigen::MatrixXf  covariance = centered.adjoint() * centered;
Eigen::SelfAdjointEigenSolver<Eigen::MatrixXf> pca(covariance, Eigen::ComputeEigenvectors);

I assumed that I get after the PCA a 2x2 matrix, but here I get then a 10x10 matrix.
Unfortunately I didn´t find another tutorials/examples, where I get the 2 Eigenvectors for a 2D point scene.

Your data matrix is stored the other way round compared to the code you borrowed, so you have to adapt the code to your storage order (and use a Matrix2f for speed):

Code: Select all

Eigen::Matrix2Xf centered    = matrix.colwise() - matrix.rowwise().mean();
Eigen::Matrix2f  covariance = centered * centered.adjoint();
Eigen::SelfAdjointEigenSolver<Eigen::Matrix2f> pca(covariance);