Using a guess transformation in GICP alignment leads to wrong results.

[saarnold]

Using a guess transformation, other than the identity, for the alignment with GICP leads to a wrong results.

It is very easy to test, by transforming the target pointcloud somewhere and using the same transformation as guess. This should lead to the same result, but it doesn't.

You can find a corresponding test case in my clone of PCL.

[VictorLamoine]

Related: issue #818

[VictorLamoine]

Appart from the fact that you have to use the inverse of the transformation as a guess; you are right.
Please test my pull request and comment if it seems right to you.

[saarnold]

Sry for the delay, I didn't had the time to check your commit until now.

In my test case I was pre-transforming the target pointcloud, in that case the same transformation has to be used as guess, not its inverse. Since the input (or source) pointcloud will be aligned to the fixed target pointcloud.

Your commit transforms the output pointcloud at the end of the optimization, which is a fair thing to do, but doesn't solve the mentioned problem.
It is still not possible to use a guess transformation, apart from the identity, in the GICP optimization process.

[SergioRAgostinho]

Closed with #989