对FPFH的问答

> Hello everyone,

>

> pcl::PointNormal contains informationabout the normal vector, the curvature and the xyz data of a point.

> Is it possible to pass to a FPFHfeature estimator the following input dataset?:

>pcl::FPFHEstimationOMP<pcl::PointNormal,pcl::PointNormal,pcl::FPFHSignature33> fpfhEst; instead of pcl::FPFHEstimationOMP<pcl::PointXYZ,pcl::Normal,pcl::FPFHSignature33> fpfhEst?

Yes, but only XYZ will be exploited for thefirst template and only

normal data will be used from the second.

> Will it still result in a good result?

It won't affect the results, you can giveit a try.

>

> What are good keypoints for FPFHfeatures?

> good FPFH features are calculated atpoints with a high curvature, so it would be good to calculate AGASTKeypoints whose are the corners in the frame, am i right?

You can calculate FPFH for any keypointsince they are disjoint processes. FPFH is a statistical descriptorthat computes histograms so question if Agast points are good keypointscan only be answered empirically since there is no directrelationship between the vicinity information (used in FPFH) and theintensity information (used in AGAST).I would rather suggestHarrisKeypoint3D which rely on geometric variation of the neighborhood to detectkeypoints the more variation you have the more your accumulated histogramwill look sharper and unique.

> Thanks for any help,