Calibration wizard Triangulation Error color

rkoma · November 10, 2025, 9:28am

General Information

Product: X36-5FA-6/20/2400-400/2400
Operating System: Linux
Data Attachments?
- Screenshots

Problem / Question

We’ve been using `X36-5FA-6/20/2400-400/2400` in our system for past 6 months and recently we experienced increased errors so we checked triangulation error in calibration wizard before conducting re-calibration.

We got green Triangulation error `0.98 mm` as you can see the attached picture.

I suppose the green color means good. But when I compared the current pointcloud with the one from 6 month ago, I can see the 10 mm offset at 2.3m away from the camera.

Also, I can see a few pixel drift from right camera between 6 month ago and now, which suggests stereo rig is loose.

My question is that the green color doesn’t mean good results?

How to verify it if the system is in good condition if it shows green color even with loosed stereo rig?

Also, how far I should put calibration board? I imagine if I put calibration board closer, the error will be decreased.

daniel.saier · November 10, 2025, 10:26am

Hi,

you can find more information about measuring the calibration in the manual: Measuring Calibration Accuracy — Guides - Ensenso SDK 4.2.1821 Documentation

I suppose the green color means good.

Yes, green means good. The thresholds for it are explained in the guide linked above.

But when I compared the current pointcloud with the one from 6 month ago, I can see the 10 mm offset at 2.3m away from the camera.

How exactly did you measure this? The error does not show in the measurement of the intrinsic geometry in your screenshot, which looks ok.

Note that you only measured the intrinsic calibration of the stereo camera. Did you calibrate the position of the camera and did you try to measure that as well. If not, is it possible that the whole camera has moved relative to the scene?

Also, I can see a few pixel drift from right camera between 6 month ago and now, which suggests stereo rig is loose.

It is very unlikely that something within the stereo camera assembly is loose and you cannot observe this by measuring the intrinsic calibration as you did in your screenshot. The intrinsic calibration is very sensitive and any moving parts of the camera invalidate it very quickly. This would produce a measurable error in one of the error metrics in your screenshot. I think it is more likely that something outside of the camera has moved.

Note that the position of the pixels in the rectified (not raw) images is not fixed in general. They can move due to a dynamic recalibration without affecting the position of the resulting 3D data. As explained above, I don’t think this is the problem in your case, though.

Also, how far I should put calibration board? I imagine if I put calibration board closer, the error will be decreased.

This depends. Putting the calibration target closer to the camera does decrease the absolute errors (at least the ones in mm). The coloring in NxView takes this into account by comparing the error to the view field at the distance of the target.

But when the camera has a large view field (like yours) it can help with the accuracy of the calibration target detection. If you have one of the 48mm calibration targets, it probably appears very small at this distance, which can also introduce noise. We recently introduced the 96mm target to help with this.

rainer.voigt · November 10, 2025, 11:05am

Actually the 6mm models use adjustable lenses, which differ from the standard configurations. @rkoma maybe you can you share the exact camera serial number?

rkoma · November 10, 2025, 1:08pm

Thank you for the link.

How exactly did you measure this? The error does not show in the measurement of the intrinsic geometry in your screenshot, which looks ok.

We have ply file from 6 month ago. So we compared pointcloud using CloudCompare.

Note that you only measured the intrinsic calibration of the stereo camera. Did you calibrate the position of the camera and did you try to measure that as well. If not, is it possible that the whole camera has moved relative to the scene?

We compare pointcloud based on the camera coordinate system. Also, I compared right and left images between 6 month ago and recent one, and I confirmed that there is no difference in left camera but I can see a few pixels drift on right camera image, which suggests the stereo rig might be loose. (And we adjusted in vergence angle by ourselves)

But when the camera has a large view field (like yours) it can help with the accuracy of the calibration target detection. If you have one of the 48mm calibration targets, it probably appears very small at this distance, which can also introduce noise. We recently introduced the 96mm target to help with this.

Thank you for the information. We also couldn’t detect board if it’s placed farther than around 2m.

Actually the 6mm models use adjustable lenses, which differ from the standard configurations. @rkoma maybe you can you share the exact camera serial number?

Let me check serial number but we needed large scanning volume so we bought lenses separately and we adjusted vergence angle by ourselves so stereo rig can be loose due to our poor assembling skill.

But my main concern is that error we got is a lot bigger than error reported in the tool. It might be our fault that stereo rig got loose but I want to know how to verify that stereo rig has problem by calibration wizard. Or it is not main scope of tools?

rainer.voigt · November 10, 2025, 1:34pm

The problem likely not your assembly skill but usually the internal lens mechanics: in our standard models we use special lens designs with a static optics group inside. Most of the standard lenses with adjustable focus can be locked with a few screws, but even tightening the fixation screws does not prevent the internal mechanics from changing slightly, and thus decreasing the calibration accuracy.

We do have C-mount lenses with 5mm and 8mm, with a static optics group and locking nuts for focusing and fixation, which remain long-term stable in our experience. What is the working volume you need, maybe one of them will fit?

rkoma · November 11, 2025, 1:35am

Thank you for your reply. I think serial number is 246108. I will check how we bought this sensor but 6mm lens is not your official product?

rkoma · November 11, 2025, 1:54am

We bought the sensor from iDS. I’m not sure the relationship between iDS and Optonic but 6mm model is not official product in Optonic?

I heard from my coworker that we selected this model from https://www.optonic.com/en/support/selector/model/?id=X36-5CP-6%2F18%2F2000-400%2F2000 in optonic camera selector.

Also, can you advice us how to select x36 with special lens designs with a static optics group inside?

rkoma · November 11, 2025, 2:08am

We recently introduced the 96mm target to help with this.

We noticed the datasheet specify 96mm target for most of the camera. Is 96mm target recommended for future calibration?

Also, why 96mm grid spacing with 7 x 7 grid becomes 390 x 390 x 4 mm dimension.

Here is our calibration board. is it too small for x36?

rainer.voigt · November 13, 2025, 1:09pm

Hi @rkoma, the 6mm variant is indeed a bit of a customization, although the camera selector will show the model specs, if you enter them directly. But it will not return it via the search function. Anyway we have to find a suitable solution. Can you maybe give us some details of the required working volume, distance, temperature and vibration conditions? And then we will suggest you a configuration with static lenses. You can also send me a PM, so we can take the discussion offline.

Regarding the measurement of the error: the calibration plate can detect most errors, but not all. Especially for large viewfields and distances it is difficult to detect non-linearities of the measurement volume if the plate does only cover a rather small part of the image, because also the plate position is unknown and will be estimated in the process of the measurement.

If you have the possibility to print a custom calibration plate (a larger one), this would reveal non-linearities much more in the measurement process. There is the NxCalTab tool in the SDK to generate print data for different sizes, above the official pattern sizes.

rkoma · November 14, 2025, 2:36am

Thanks, I will send a PM to you

rainer.voigt · November 20, 2025, 8:44am

Hi Ren,

for completeness of the forum thread:

Suggested alternatives would be:

As a simple lens exchange on your current X-series:
- 8mm static lenses, like X36-5CP-8/16/2750-400/2750. This would give you the required volume between ~2m to 4m distance.
And for the similar volume there are corresponding, current C-Series models:
- The 8mm model with focus distance set to 3m distance C57-8/3000-M will give you the 1.5x1.5m view field at around 2.3m - 4.3m distance, similar to the 8mm X-series
- And a 6mm option: C57-6/2500-M, which hat the intended working volume at around 1.9m - 3.9m.

I sent you the PDF of the old PCN from 2021 for the X-series lens change by mail. Current and future PCNs will be published here in the Compliance category, you can simply subscribe to the entire category if you want to get email notifications for all changes.

Best regards, Rainer

rkoma · November 20, 2025, 9:38am

Thank you so much for the support.

rkoma · November 20, 2025, 9:59am

Let me add one thing. It’s important to ask if the model use static lenses before buying. Some models use customized lenses which is not suitable for long-term usages.