About the parameters in different environments

[JixinGao]

Hello! Thanks for this great work!
I'm trying to use it for ground segmentation of point clouds in an indoor environment, which is relatively narrow.
I have adjusted some parameters, but the segmentation is not as accurate as it is outside. There are many error points that are segmented as ground points, including ceilings, objects on the ground, and a few vertical walls. Would you give some suggestions on parameter adjustment？
The picture is as follows (The red points are the ground points segmented by patchwork++):

[LimHyungTae]

I guess ground segmentation failed. Could you give me some pcd files?

[JixinGao]

Hello, thanks for your reply, I embedded patchwork++ into SLAM, this is a .pcd file of the mapping results (10Hz point cloud measurements) for 20 seconds (I do ground segmentation of each frame of point clouds).
scans.zip

[LimHyungTae]

That's weird. In Patchwork++,

1. it assumes that LiDAR coordinates should be front-left-up coordinates
2. Is your system handheld?
3. Which parameters did you change? Segmenting ceilings and tripods does not make sense.

[JixinGao]

It's not handheld, it's placed normally in dataset.
This is the parameters (there are a few variables that I add additionally for other functions).

Could it be that is caused by the pcd is the SLAM mapping results? I use patchwork++ to detect the raw LiDAR data of each frame. However, there is no rotation between the mapping coordinate system and the Velodyne coordinate system. This is the mapping in rviz:

[LimHyungTae]

I saw there are a few ground points. Could you change the sensor height and the grid size?

[JixinGao]

In fact, the data used comes from the dataset (this is in the initial phase, so the nearby ground is blind for LiDAR), but I will soon conduct real-world testing, I will follow your advice and make the sensor detect more ground points.
By the way, do you think that the few number of ground points is what caused the poor segmentation result? Does the parameter file need to be modified?
Thank you for your reply !

[LimHyungTae]

It might be. Going to the your problem back, I don't think ground segmentation is needed in your case since in that case, ground points are not dominant, aren't they?

[JixinGao]

Thanks for your response. Indeed, there are few ground points in this scenario, but they are still important to me. I will adjust the LiDAR FoV to observe more ground points, and thank you again for your help!