Correspondence grouping can't find instances

classic Classic list List threaded Threaded
2 messages Options
psi
Reply | Threaded
Open this post in threaded view
|

Correspondence grouping can't find instances

psi
This post has NOT been accepted by the mailing list yet.
Hi, all!
I've tested Correspondece Grouping algorithm on own data. But it was unsuccessful. I have a couple of questions:
1) As I understood, this algorithm works with different orientations of the model w.r.t the object, right? I mean, it does not matter what is the orientation of model.
2) The scale (or size) of the model must be the same as of object, yes?
Reply | Threaded
Open this post in threaded view
|

Re: Correspondence grouping can't find instances

pclbel
This post has NOT been accepted by the mailing list yet.
@psi I was doing a lot of tests with correspondence grouping in the last weeks. What I can say from now is that the algorithm is very sensitive to the paramters that are passed in combination with your model and scene.

to Point 1) I tried to match a model against itself, but only translated it a little bit (so my scene is the translated model). The result was that the number of correspondences found get less and thus at some point no instances are found anymore (No change in orientation, only position!).
I found out, that the normals can and will usually flip and therefore will be different. The reason is that the normals are computed dependent on the viewpoint, so different orientaion or even position results in different normals. I think what happens then, is that the correspondence matcher cant find anything because the normals orientation is ambigious. So whith every differnece in the 6DOF of the model vs. scene the normals can be flipped and you would have to change the viewpoint just before you calculate the normals.

to Point 2) I would be happy it would work with same scale for the first