FPFH formula different then the Rosu paper?

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

FPFH formula different then the Rosu paper?

nyarbel
I have dug into the source code in order to uderstand exactly how FPFH with a search surface works in case to keypoints are not exactly from the surface. Doing it I seem to have stumbled upon a difference between the feature as described in "Fast Point Feature Histograms (FPFH) for 3D Registration" and the PCL implementation.
FPFH Formula
A.It seems that the first term(SPF of source point) is absent from your calculation
B. the 1/k factor is absent, instead, the histogram has been modified to sum to 100 along each angle 11 bin.

Can you please confirm my assesment?
anyone know what are the pros/cons of this?