Question for ICP registration

[ha485]

Hi, After ICP registration, I've got the result below

1 .000000 -0.000111 -0.000344 0.064012
0.000111 1.000000 -0.000059 -0.052990
0.000344 0.000059 1.000000 -0.192858
0.000000 0.000000 0.000000 1.000000

That means x,y,z are moved 0.064012, -0.052990, -0.192858, respectively. but The value is overestimated.
when I compared before and after registration, the difference between centers was x= -0.005, y= -0.003, z=-0.0121.
I don't understand why the translated values are so overestimated.

[daniel]

Mind that the translation is applied after the rotation (P' = R.P + T). This means that the translation also compensates for the rotation (which implicitly happens about the origin point (0, 0, 0)).

[ha485]

Rotation values are almost zero in the transformation matrix.
What I'm curious about is the last column values 0.064012, -0.052990, -0.192858 are too high.
It doesn't make sense at all because the actual translation movement is small (x= -0.005, y= -0.003, z=-0.0121).

[daniel]

Assuming the rotation is small enough to not cause any shift of the translation component, then maybe you have to work on the ICP initialization and parameters? Are the cloud initially roughly registered? Have you considered the overlap between the two clouds?

[ha485]

Could you have a look my pointclouds?
If you want, I can share the pointclouds.

[daniel]

Yes, don't hesitate. Send the link to admin@cloudcompare.org