FPFH implementation does not match documentation

⚠️ This is a issue tracker, please use our mailing list for questions: www.pcl-users.org. ⚠️

The FPFH equation:
FPFH(p) = SPFH(p) + weighted average of SPFH over neighbourhood
does not hold for the current implementation.
PCL FPFH consists only of the weighted average, the query point SPFH is omitted.

This can be seen from the code logic, in features/impl/fpfh.hpp:

1. computeFeature calls computeSPFHSignatures and then iterates over points with weightPointSPFHSignature
2. computeSPFHSignatures fills three eigen matrices with SPFH split by feature
3. weightPointSPFHSignature clears the FPFH, then iterates over neighbours, skipping the query point and adding weighted neighbour SPFH to the FPFH, and finally normalizes FPFH.

If this is intentional, I guess the documentation, tutorial and the wiki page should mention it.

Your Environment

• Operating System and version:
• Compiler:
• PCL Version:

Expected Behavior

Current Behavior

Possible Solution

Code to Reproduce

Context

Start in features/impl/fpfh.hpp and trace computeFeature, computeSPFHSignatures, and weightPointSPFHSignature. Compare the implementation with the documented FPFH equation, then determine whether omitting the query-point SPFH is intentional. Done means resolving the discrepancy in the implementation or updating the documentation, tutorial, and wiki consistently.