DLT
only works when the transform function is linear, doesn’t deal well with outliers.
match feature points
노이즈가 많은 이미지에서 정확도가 떨어질 수

Solving
Given , solve for H such that
- For each correspondence, create 2x9 matrix
- Concatenate into single 2n x 9 matrix
- Compute SVD of
- Store singular vector of the smallest singular value
- Reshape to get
RANSAC loop
make good to outliers
- Get four points correspondences (randomly)
- Compute H using DLT
- Count inliers
- Keep H if largest number of inliers