What is the cause of Z-axis resolution artifacts for structured light 3D cameras?

[pauljurczak]

I’m capturing point clouds with Orbbec Astra Mini structured light 3D camera (similar to this one https://www.orbbec.com/products/structu ... -mini-pro/) and I noticed that Z value resolution is pretty coarse at longer distances. For example, at ~2.6m distance to a flat target, the resolution is about 28mm, i.e. I’m getting these points:

[-123.9576 -187.60762 2682. ]
[-118.79163 -187.60762 2682. ]
[-113.625656 -187.60762 2682. ]
[-108.45967 -187.60762 2682. ]
[ 51.146034 -185.649 2654. ]
[ 56.25808 -185.649 2654. ]
[ 61.37012 -185.649 2654. ]
[ 66.48217 -185.649 2654. ]

and no points with Z coordinates between 2654 and 2682. Is this normal? Why Z values are integral with gaps increasing proportionally to target distance. I understand why the accuracy will decrease with target distance, but integral values are surprising. Do you get similar artifacts with other 3D cameras?

Cross-posted from https://3dclub.orbbec3d.com/t/astra-min ... value/4596 and https://robotics.stackexchange.com/posts/118163.

[MJ17]

What you’re seeing is most likely normal for this type of camera

Astra Mini depth is usually output as integer millimeters, but the bigger issue is that the camera does not measure Z directly. Structured-light depth is computed from disparity, and after converting disparity to Z, the depth steps get larger as distance increases. So at approx. 2.6 m it is expected to see quantized “bands” in Z, with missing values between valid levels

So yes, similar artifacts can appear on other structured-light and stereo 3D cameras too, especially at longer range and on flat surfaces. It is not just lower accuracy - it is also coarser depth quantization. If you need smoother Z, try shorter range, temporal averaging, or a different sensor technology