SURF detector producing too many results

[Mark W]

I would like to reduce the number of results the SURF feature detector is producing. At the moment it is detecting more than 7000 features per frame. This is an unmanageable number in my environment. I have played with the octave and and interval parameters. Reducing these nets 0 features. Increasing them only makes the problem worse.

I have better control with the opencv SURF detector. OpenCV accepts a hessian minimum threshold parameter. If I increase this, I receive fewer, but arguably stronger features. Has anyone had any success with implementing similar filtering in libmv? I have looked myself, but I have to concede that my understanding of SURF feature detecting is not sufficient to implement a solution.

I appreciate any help on this issue.

[Sergey Sharybin]

Hi,

Those sets of detectors is quite dusty now (at least to my knowledge), needs some revisit and tests. Keir will know their status better tho.

The only detectors which i'm sure about are implemented in simple_pipeline/detect.cc. In fact, those are the detectors which are used by Blender. Perhaps they might be handy for you. 

As for SURF, surely we have to look into them, but i can't give time estimate for that at this moment.

--
--
========================
libmv...@googlegroups.com
http://code.google.com/p/libmv

---
You received this message because you are subscribed to the Google Groups "libmv-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to libmv-devel...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

With best regards, Sergey Sharybin