Added H decomposition #121
Conversation
The code was originally written by @yaqding Co-authored-by: <yaqding@users.noreply.github.com>
| Eigen::Matrix3d V2 = svd.matrixV(); | ||
|
|
||
| if (abs(S2(0) - S2(2)) < 1.0e-6) { | ||
| poses->emplace_back(H2, Eigen::Vector3d(0.0, 0.0, 0.0)); |
There was a problem hiding this comment.
Should maybe also add a zero vector to normals here.
Also this check could be made invariant w.r.t. the scale of HH.
There was a problem hiding this comment.
So maybe scale this w.r.t. the determinant?
There was a problem hiding this comment.
Could divide by S2(0) as well.
There was a problem hiding this comment.
I think it's best to have poses.size() == normals.size() even in these corner cases. So normals->emplace_back(0,0,0) is reasonable here as well I think.
| @@ -19,6 +20,10 @@ focals_from_fundamental_iterative(const Eigen::Matrix3d &F, const Camera &camera | |||
| const int &max_iters = 50, | |||
| const Eigen::Vector4d &weights = Eigen::Vector4d(5.0e-4, 1.0, 5.0e-4, 1.0)); | |||
|
|
|||
| // Estimate the camera motion from homography. | |||
| // If you use H obtained using correspondences in image coordinates from two cameras you need to input K2^-1 * H * K1. | |||
| void motion_from_homography(Eigen::Matrix3d HH, std::vector<CameraPose> *poses, std::vector<Eigen::Vector3d> *normals); | |||
There was a problem hiding this comment.
HH could be const ref unless we need to copy.
There was a problem hiding this comment.
This is done so we can enforce the determinant to be positive inside the function.
| } | ||
| Eigen::Vector3d t2 = (H2 - R2) * n2; | ||
|
|
||
| poses->emplace_back(R1, t1); |
There was a problem hiding this comment.
I guess there are four possible solutions here
(R1, t1, n1)
(R1, -t1, -n1)
(R2, t2, n2)
(R2, -t2, -n2)
So here we choose the solution pairs with n(2) > 0, but is it possible to decide that these are the correct ones without any correspondences?
There was a problem hiding this comment.
Might be better to return all four otherwise.
|
I forgot to comment. The requested changes have been committed. Let me know if this is ok so I can also add changes to README. |
|
Looks good to me, thanks! Please fix the normals thing in the comment, and then update the README and we can merge. |
* add H decomp description to README
|
I added the normals output and description in README. Should I also bump the version? I noticed there have been some other somewhat significant commits without version change since 2.0.4. |
Added homography decomposition into pose using SVD.
Original code by @yaqding I just changed the call signature and added bindings.
If approved I will add info to README and bump version.