Transformations bring mistakes in estimation

[zinuoyi]

Hi,

thank you for this great work. But I would like to share an interesting thing:
It cannot handle the ground correctly if we transform the points from the ego frame to the world frame.
(Left: transformed; Right: not transformed)

You can see there is a red circle in the left pic.

Is this supposed to be so, like by design we should transform the points to the ego frame first?
PS. Do you think it is possible to improve this since it sounds reasonable that a tranform of a scan on the x-y plane should not make a difference in the ground removal.

Best,
Zinuo

[LimHyungTae]

It may affect the changes of z values. Btw, why don't you run ground segmentation first before transformation?

[Navsea]

It looks like this is related to the Z0 zone, the zone that patchwork ignores as it assumes it will be empty.
(sensor does not look directly down, vehicle would block rays anyway)
The red circle (non ground) lies more to the rear, as you have shifted your point cloud forward (I assume you're vehicle ref frame lies behind the sensor ref frame).
You could easily confirm this by changing the minimum ring distance, see param "min_range".

I guess the solution would be to redefine your CZM so that the origin of the model lies on top of the sensor origin expressed in vehicle frame.