Fusion stopped working correct after few moths

Dear Stereolabs,

we have been using the Fusion setup with 2 cameras for some months now and since a few days even after a clean calibration the fusion starts doing wild errors.

Is it possible that the back wall of the Installation (pure White) is a problem for the cameras?
The backdrop gets direct sunlight since a few days or weeks and since then the fusion is creeping around very badly.

Is there any way we can fix this, without changing the installation?

the installation is in a museum and therefore its important that it runns stable.

This is a possible cause of the problem. Can you share pictures showing this particular condition?

Hi,

Did you change anything else? The ZED SDK version, for example?

Thanks for your quick answer :folded_hands:

I made you pictures.

  • since the morning the direct sunlight is now gone, however the fusion still creeps around just differently.

the images:

im 100% certain this is something new, because we had it running and it worked so well.
you can see this video shows how it was before: https://www.vrfx.ch/files/videos/mdxbcf74/Digital-Wardrobe-In-Action-small26c9f3c4-9967-4103-ab6d-ef761d103932.mp4

@pachermann, Can you provide the information requested by my colleague @BenjaminV?

(post deleted by author)

No we did not change anything on the Software, nor SDK nor did we compile anything new, it just came up like this one day suddently.

it is a finished installation that runns since a few months.

You said you did a new calibration. Could you try the old one ?

From what I see in the video you shared, only the fused skeleton is completely broken; the “raw” ones are ok.

I have overwritten the old one because it behaved exactly the same way.

it went totally crazy and cloned some of the skelettons, it basically looked the same way more or less.

that is why i attemped to re calibrate becuase we though visitors have move the cams.

What does that mean, how can i solve this?

(post deleted by author)

After looking a your video again, I noticed the fused skeleton is most likely the purple one, not the red one. The red skeleton might come from a false detection, which may explain why the result is that bad.

You can try to avoid this from happening by increasing the detection confidence threshold (BodyTrackingRuntimeParameters Struct Reference | API Reference | Stereolabs) so that low-confidence detections are filtered out.

Thank you @BenjaminV so is this been done in the fusion c++ file? the one i make the exe for the fusion server?

Yes, here to be precise.