Items with no label
3335 Discussions

D415 depth noise at fairly short distance

JLafe
Beginner
1,536 Views

Hi,

In the context of our application, the main object of the scene is located between 25 and 34 cm from the camera. I did a setup with a flat plane (a glass that has been painted) and I observe a circular depth noise. The noise is definitely not related to the plane itself. If I move forward or backward my plane, the noise stays at the same place in the image and doesn't follow the movement of the plane. If I change the distance of my flat plane by a couple of cm, the noise get slightly modified but still have similar shape. The maximum depth variations I've quickly measured are 3.5mm at a distance of 35cm, 2.4mm at a distance of 28cm and 1.4mm at a distance of 25cm. I understand that these values respect the spec of 1% precision but since this noise can be characterized and is not simply a white noise, I'm looking to minimize it via acquisition settings. Since there are now a lot of advanced setting, I'm looking for the ones that could help me on that matter.

Attached are images of the noise with the flat plane located at different distances. Even tough the camera plane and the surface plane should be parallel, there might be a small slope but we still can clearly see the noise shape.

Thanks

0 Kudos
7 Replies
MartyG
Honored Contributor III
277 Views

If the glass has reflective qualities even if painted then this can affect the depth quality. This noise may be intensified if you have bright lighting in the room. Reflectiveness can be dampened for better image results by spraying a fine powder such as foot powder or baby powder onto the surface.

In regard to depth accuracy, there is a relationship between accuracy and the resolution you are using. The lower the resolution that you scan at, the grater the depth error that can be introduced. I believe that a minimum resolution of 848x480 is recommended for good accuracy, with higher resolution (1280x720) being better.

With black borders like the one in your image, I usually recommend de-selecting the 'Auto Exposure' option in the RealSense Viewer software and changing the Exposure and Gain sliders manually until you get less noise in the image.

0 Kudos
JLafe
Beginner
277 Views

Hi Marty,

Thanks for the quick follow-up. Unfortunately, I see the same noise shape and intensity why I use a cardboard as the flat surface; I don't like using cardboard since its flatness is not that great, but since it does show the exact same noise, I must acknowledge that it is flat enough. That lead me to say that this noise is not related to reflectiveness. I also closed all the lights in the room and I still see the same noise shape and intensity.

As for the resolution, at 25 cm I did not measured significant difference between 640 x 360 and 1280 x 720 on the noise shape and intensity. I did saw a difference on the noise intensity when I moved my surface further such as a distance of 34 cm.

I also removed the auto-exposure and changed manually the exposure as well as the gain, but the noise shape and intensity are still there and do not seem to be really affected by these settings.

Is there any advanced settings that are controlling the noise compensation or that have a direct effect on the noise and its shape?

Thanks

0 Kudos
MartyG
Honored Contributor III
277 Views

The next thing to try to reduce disruption would be to reduce the Laser Power slider setting, which is under the Controls section of the RealSense Viewer options.

0 Kudos
JLafe
Beginner
277 Views

I already tried that. It does indeed affect the noise. By reducing the emitter power or even turning it off (with a flat surface that has texture), the noise shape is more smooth. Actually, the images I previously posted where with the emitter enabled set to off. By turning it on with a strong power I get a grid effect on top of the noise I already described. See attached image for the difference.

Anythings else I could try?

Thanks

0 Kudos
MartyG
Honored Contributor III
277 Views

The IR emitter projects a pattern of dots onto what the camera is observing. Reducing laser power reduces the visibility of the dot pattern, as you observed.

Does the room that the camera is in have florescent lights like ceiling strip lights? These can create more noise than bulb lights because of a flicker rate that is hard for the human eye to see.

0 Kudos
JLafe
Beginner
277 Views

Yes, our lights are florescent lights but even after closing them and use the emitter as the lowest power as possible to still see something, I do see the same noise pattern and intensity.

Should I compensate the noise with some post-processing or is there any advanced settings that directly affect the noise shape and intensity?

Thanks

0 Kudos
MartyG
Honored Contributor III
277 Views

In cases where there is noise that you cannot get rid of, doing a Point Cloud scan instead of a conventional depth scan can provide a tidier image. You can view the data as a point cloud in RealSense Viewer by clicking on the '3D' option in the top right side of the window.

I also recall an old trick about turning the camera side on or upside down so the projection light falls on the object differently.

You may also get more detail if you put a background sheet / board behind the object to act as a horizon for the image ,or surround the object with other objects to give the camera more depth points to lock on to.

0 Kudos
Reply