7_fine_co-registration.md

Fine co-registration with MSA

After the coarse co-registration, the point clouds are matching but there's still an error on this. To improve the co-registration (reduce this error), we use the MSA algorithm which uses plane patches to match the different scans more closely.

This can be done after one or multiple scan positions are coarsely registered. The next parts are based on doing the fine registration after one scan position has been coarsely registered.

This shows you an example of the error which may be left after coarse co-registration. See how the point clouds of the two scan positions (white verus blue color) are not matching.

Steps

1. Click Registration.
2. Click Multi Station Adjustment.
3. Click Start adjustment....

4. Make sure the registered scan positions are activated by selecting them and right-click.
5. Click Activate. A check symbol will apear left next to the scan.
6. Make sure the other scan positions which haven't been registered yet are deactivated. If they are activated (checked) do step 1 and click Deactivate.

7. Make sure the registered scan positions which have already gone through fine co-registration are locked by selecting them and right-click.
8. Click on Lock position and orientation. Froozen scan positions (such as ScanPos001) are automatically locked.

9. Make sure the scan position which you want to finely co-register is unlocked by selecting it and right clicking.
10. Click on Unlock position and orientation.

11. Under INPUT DATA make sure Use plane patches is indicated. Ignore the tiepoint, tieobjects and measured scan positions.
12. Under PARAMETERS the most important one is search radius (which you should choose as the distance you think the error between the two scans is approximately). The other parameters can generally stay unchanged.

More information on the other parameters can be found on the help page of RiSCAN PRO:

13. To start the adjustment, you set your parameters based on the error you observe.
14. Click Calculate.

It can be good to visualise your point clouds (ScanPos001 and ScanPos002 in this case) to see what MSA is doing and if the adjustment is improving the co-registration between scans.

15. The numbers under ΔX, ΔY, ΔZ, ΔRoll etc. show how much the SOP changed (the point cloud translated and rotated).
16. Under STATISTICS, the Error (StdDev) [m] also gives an idea of the resulting error.

17. Generally an Error (StdDev) [m] </= 1 cm is good. So you can reduce the Search radius to approximately 2 times the error.
18. Click Calculate.
19. Repeat the process untill the co-registration looks good/doesn't really change anymore.

20. Now the error is below 1 cm. This scan position is finely co-registered now.
21. Right-click the scan position.
22. Click Lock position and orientation.
23. IF something went wrong during the MSA adjustment you can undo an adjustment by clicking Undo last (which undoes the last adjustment) or Undo all (which undoes everything you did since you opened the MSA screen).

Now you repeat the coarse and fine registration steps with the next scan position untill all the scan positions are finely co-registered.

Final fine co-registration

Because of this sequential process errors can propagate towards the end of the scan positions. Therefore it's advised to run MSA algorithm once more on all the scan positions.

24. Open MSA (see steps 1-3).
25. Activate all scan positions (see steps 4-5).
26. Unlock all scan positions (see steps 9-10) expect for ScanPos001 which is frozen.
27. Set the Search radius [m] to a low number such as 5 cm.
28. Click Calculate.

It is important to visually check the results of the final co-registration.

29. Visualise all point clouds in different colors (see previous steps).
30. Set Top view.
31. Set Orthogonal view.
32. Rotate the point clouds so they are aligned with the X-Y axis in the grid.
33. Click Selection mode > Shape > Rectangle selection.
34. Select the plot based on the scan positions you see.

35. Click Show only selected area.
36. Click on the Height filter.

37. Use the Height filter to take slices through the plot in both X, Y and Z.
38. Go through the plot and check the point clouds of different scan positions (in different colors) are matching.
39. If something is off try to correct it using MSA. (see below to know if something is "correctable".)

When is something bad registration (and can be fixed) and when is it just wind (and cannot be fixed)?

When you see ghosting effects (an object is in the point cloud multiple times on a slightly different position in different scans) that are mostly present in the branches but not in the stems, with the effect bigger towards the top, this is most likely due to wind. Leaves can easily be moved by wind, therefore we focus on stems to check wether the co-registration is OK.

You can figure out which scan positions are showing this ghosting by looking at the point clouds, hiding some scan positions (in the object inspector) and seeing if the ghosting effect disappears.