Items with no label
3335 Discussions

Precision in depth-map

BHerl1
Beginner
2,011 Views

I am aware of the rs-depth-quality tool, but I wonder if work similar to e.g. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3304120/ this paper has been done with the Intel D400-cameras.

It would be useful to have methods in the SDK to get theoretical precision of depth-estimation for a given depth, maybe even estimated precision for each pixel in "current" depthmap. Such calculations would be subject to most parameter-settings in the camera, external conditions like e.g. light, as well as details of algorithms in the ASIC. It would therefore be hard to calculate this in applications. Any plans or work in this area?

Regards,

- Bjarne

0 Kudos
1 Solution
idata
Employee
356 Views

Hello BjarneG,

 

 

Thank you for your feedback. We will report this to the RealSense development team.

 

 

In the meantime, it may help you to know that we have done precision tests on the depth accuracy of the D400 cameras and have found that the mean error of the depth accuracy is less than 1% (around .5%) of the distance from the camera. This means that at a distance of 2meters the error in the reported distance can be around 1 centimeter (these results are not published). Does this help?

 

 

Regards,

 

Jesus

 

Intel Customer Support

 

View solution in original post

0 Kudos
4 Replies
idata
Employee
357 Views

Hello BjarneG,

 

 

Thank you for your feedback. We will report this to the RealSense development team.

 

 

In the meantime, it may help you to know that we have done precision tests on the depth accuracy of the D400 cameras and have found that the mean error of the depth accuracy is less than 1% (around .5%) of the distance from the camera. This means that at a distance of 2meters the error in the reported distance can be around 1 centimeter (these results are not published). Does this help?

 

 

Regards,

 

Jesus

 

Intel Customer Support

 

0 Kudos
BHerl1
Beginner
356 Views

Thank you for a quick reply Jesus!

Yes, this is informative. But I would be very surprised if this is not depending on the active settings in the camera, e.g. whether it is set to "High Accuracy" or "High Density". (I would also expect it to depend on light-conditions, but first and foremost on settings.) Any further information on this?

Best regards,

- Bjarne

0 Kudos
idata
Employee
356 Views

BjarneG,

 

 

You are correct, this is highly subject to environmental factors and camera settings. I don't have further information on how these tests were performed other than 80%ROI, AE, 60mW laser power, 1280x720, 30FPS.

 

 

Results will vary widely depending on the many factors that influence performance. It is always best to test and tweak parameters for your particular situation.

 

 

Jesus
0 Kudos
BHerl1
Beginner
356 Views

Thanks for details Jesus.

It is not always possible to test in the right environmental conditions (and these may also vary wildly), hence some study like the mentioned paper may serve as a guideline for practical use. It would, for example, be valuable to see how much precision changes with various presets in the same environment.

I look forward to further information from your team of how to approach this topic.

Regards,

- Bjarne

0 Kudos
Reply