How to limit the degrees of freedom of the point cloud registration

[diarmaidocualain]

Hi!
I have a camera that is generating point cloud data. The camera has a fixed position, but is free to roll roll, pitch and yaw in any direction (up to certain degrees). It captures new point cloud data at each new orientation. Each new orientation is unknown.

The point cloud algorithm that I am using (Super4pcs) outputs the rotation and a translation values. Due to my camera having a fixed position, there should not be a translation output. I have confirmed this in experimentation: The three x,y, & z translation outputs of the algorithm are very small while giving an accurate values for the x, y & z rotation.

However, I would like to speed up the processing time. Due to the cameras fixed position, I should be able to remove three degrees of freedom, i.e. the translation calculation. Would anyone have any hints or suggestions of how I would go about doing this in the algorithm?

[nmellado]

Hi,
This is actually a very good question and interesting use case.
Could you share a couple of point-clouds so I can see how to extend the library to do rotation-only alignement ?

Thanks

Nicolas

[diarmaidocualain]

Yes, both point clouds are given in the zip file below:
model_and_data_point_cloud_on_unit_sphere.zip