Is the lidar tilted?
Why do all the point clouds have a roll and pitch? Can you explain?
We try to mount the lidar as “horizontal” as possible, but even vibrations can affect the extrinsics. Therefore we regularly calibrate the vehicles to measure the actual roll and pitch.
Even on a flat road the recordings are tilted. Could you estimate how much change can occur because of vibrations and other effects?
Could you provide the pictures of your mountings or give us an idea how you have mounted with a drawing maybe?
You can see a few example photos at https://www.nuscenes.org/gallery. When the dataset was collected we had a fleet of vehicles in 2 countries, and it is unavoidable that they will differ slightly. I suggest you work with the extrinsics in the dataset to see how much they vary across weeks and vehicles.
So it is also possible that the orientation can change even between frames. Correct?
The calibration parameters do not change between frames, but the actual orientation can change slightly.