Switching to squared sampson point-line error

[ayushbaid]

The squared sampson error is the first order approximation of the gold-standard reprojection error.

This is a work-in-progress PR, and I'll see if the switch actually improves performance. I will clean up the PR and add more documentation if we want to merge this.

[ProfFan]

Wait, is this a WIP PR? @akshay-krishnan @ayushbaid ?

[ProfFan]

Also this is a breaking PR.

[akshay-krishnan]

Fan, yes this a WIP PR. We merged a portion of it in another PR. I am not sure how Ayush wants to proceed with this, I think we will open a new PR if we want to merge the remaining changes.

[ProfFan]

Yeah what I mean is that this will change how old code behaves or not?

[dellaert]

It seems this was merged in develop, not in another branch, so Fan questions are pertinent :-) Phan, why do you say this is a breaking PR? CI seems to work.

[ProfFan]

@dellaert Because

double EssentialMatrix::error(const Vector3& vA, const Vector3& vB, 
                              OptionalJacobian<1, 5> HE,
                              OptionalJacobian<1, 3> HvA,
                              OptionalJacobian<1, 3> HvB) const {

signature has changed, and

*HE = 2 * sampson_error * (numerator_H / denominator -
         algebraic_error * denominator_H / (denominator * denominator));

The first Jacobian has changed as well.

[akshay-krishnan]

Fan, those changes have not been merged into develop. Here is the error() method in EssentialMatrix.cpp at develop.
Ayush, merged some commits from this branch feature/sampson-epipolar-error into feature/essential-mat-with-approx-k which is what #753 used. But I removed those changes in this commit before merging #753.

So git will say that this PR has been merged since its commits went into #753 but I removed most of the changes in those commits later on.

[ProfFan]

@akshay-krishnan I see, thanks :)