How to access hessian matrix to study observability?

[SubodhMishraIdeaForgeTech]

Hello,
I am trying to use g2o's apis to access the hessian matrix to study the observability of the system I am trying to minimize. My goal is to access the Hessian matrix and then look at their singular values. If they are under a threshold then it may be an ill conditioned problem. Is there a straight forward way to extract the Hessian matrix? I appreciate you time!

[SubodhMishraIdeaForgeTech]

I discovered that the saveHessian function may be used to dump it into a txt file but is it possible to extract it directly as a matrix or something?

[RainerKuemmerle]

One way could be to subclass from the

template <typename Traits>
class BlockSolver

to provide read access to the block matrix structures. The members are protected and should be accessible in the sub-class.

[SubodhMishraIdeaForgeTech]

Thanks! I will try this out.

[SubodhMishraIdeaForgeTech]

@RainerKuemmerle , thanks for your previous response. I was analyzing the the Hessian Matrix and I noticed that the condition numbers I get are very high, in the order of millions. Is it a stable solution or does it mean that my problem is ill posed? Does g2o provide any insights about the quality of the estimate?