Point cloud is not stable

Hi ! I have tried to measure the width of chessboard by using Zed 2. My chessboard is 8x6 at 6 meters away with the width 8cm, and it can be measured with the (mean, std) = (7.890 cm, 0.38212 cm). But when I put the chessboard at 7 meters away, I get the data is (mean, std) = (8.7189 cm, 1.3465 cm) . And the point cloud is not a plane.
Following is my setting
init_params.camera_fps = 30 # Set fps at 30
init_params.depth_mode = sl.DEPTH_MODE.ULTRA # Use PERFORMANCE depth mode
init_params.coordinate_units = sl.UNIT.METER # Use meter units (for depth measurements)
init_params.depth_stabilization = True
init_params.depth_minimum_distance = 3.0
init_params.depth_maximum_distance = 10.0
runtime_parameters = sl.RuntimeParameters()
runtime_parameters.enable_fill_mode = True

Should I need to recalibration or setting another parameter?

Hi @GU-Lin
Welcome to the Stereolabs community.

This to use NEURAL depth mode:
init_params.depth_mode = sl.DEPTH_MODE.NEURAL # Use PERFORMANCE depth mode
and raise the depth stabilization
init_params.depth_stabilization = 50

Do not enable fill mode for depth measurement tasks:
runtime_parameters.enable_fill_mode = False

Hi @Myzhar
I have tried your setting but it doesn’t get better result. Here is the point cloud image by

using meshlab at 8 meters away.

I mesure the data for width of chessboard and the result is (mean, std) = (10.845 cm, 3.12556cm). But the correct width of chessboard is 8 cm.

Can you post a picture of the chess board?

Hi @Myzhar
Here is the link. You can find the chessboard image, point cloud and a screenshot.

While taking picture, there has some periodic noise. Do you have any idea?


The chessboard is far from the camera and the texturization is not good (repetitive and poor in visual features) to extract a point cloud of the surface in such conditions.

This can be caused by the frequency of the artificial illumination.

How far is the chess board from the camera? What are the sizes?

How far is the chess board from the camera? Ans : About 8m.
What are the sizes? Ans: Each grid width is 8cm and the pattern is 80 cm x 60 cm.
But Zed 2 in spec says the depth range can achieve 20m.

The depth precision is about 1% for the measured depth value in optimal conditions, so you can expect an error of ~8 cm at 8 m of distance with a good texturized surface.
In your case, the chessboard does not provide good visual features, so the expected error can be higher than 1%.

A curiosity, did you perform a manual calibration of the camera?

No, I did not calibration the camera. Just try any parameter, like depth mode or fill mode.

Can you record an SVO in the same conditions by using ZED Explorer and share it with me?
I can test it to search for optimal configurations.

Hi @Myzhar ! Sorry too late for your response, because my expernment envienment need to be changed recently. But it still has the same problem. So I also record the SVO. Following is the chessboard expernment condition.

  1. 8cm / grid, and total pattern is 80cm x 60 cm.
  2. distance : 7m away
    Thanks for your help.
  3. URL : problem - Google Drive

The point cloud of the chessboard seems very good to me

The distance of each point is in the expected precision range which is ~2%/3% at 7 m

A curiosity: why is the light so yellow? Did you disable the auto white balance?

No, I didn’t disable the auto white balance. It is the ambient light and there is no way to avoid it. How do you set up the configurations.

Click on the gear on the top right to open the camera settings:

In Python, you can control all the settings by code.
On GitHub, we provide a useful example: https://github.com/stereolabs/zed-sdk/tree/master/camera%20control.