Distance to closest point on an object, BW1098OBC

rldeveloper

Hi!
I am working on getting familiar with the BW1098OBC unit. I am however having some trouble with getting accurate distance to an object. When running "python3 depthai_demo.py -s metaout depth -bb" (depthai v0.3) I am getting completely wrong distances to objects, e.g 2 meters to an object 50cm away.

The calibration values read out from eeprom match the default calibration values specified the BW10980BC.json file.
Does anyone have any idea on what might be wrong?

Additionally, I am wondering what point on the detected object the x,y,z values refer to? Is it possible to get the distance to the closest point on the detected object?

Brandon

Hi rldeveloper ,

Sorry about the delay. Our staff was mostly offline over the past 5 days because of Thanksgiving here in the US.

So I think the issue on the distance being incorrect is that the BW1098OBC has a minimum depth of _0.7 meters. (See here)

So we now support both extended disparity and also lower resolution operation, both of which independently halve the minimum depth. So running extended disparity at full 1280x800 resolution on the BW1098OBC would result in _0.35m (35cm) minimum depth. And runnning 640x400 resolution -and- extended disparity would result in a theoretical minimum depth of _0.175m (17.5cm) but actually would be limited by the min in-focus distance of the grayscale global shutter cameras, which is 0.196m (19.6cm).

So I'd recommend trying out extended disparity from here, and visualize the depth (like the default example does) to see if you can properly get depth information when extended disparity is enabled:
https://github.com/luxonis/depthai-experiments/tree/master/gen2-camera-demo

For example, above, you can see my knuckles are too close to that BW1092 w/ subpixel enabled.

Thoughts?

Thanks,
Brandon

Brandon

Oh also for pulling the closest depth of an object, yes, we can provide an example of how to do this. I was just discussing with Lukasz offline.

rldeveloper

Thanks a lot for explaining the minimum depth issues. I will try out the experiment you linked. And its great that you can provide an example on how do find the closest depth!

Brandon

Thanks rldeveloper , yes for sure! So we made a Github issue to track this implementation as well:
https://github.com/luxonis/depthai-experiments/issues/17

hopkira

Hi, first of all congratulations on a brilliant product - I've been playing with the OAK-D for a few hours now - I wondered if the extended disparity mode mentioned above is available on the OAK-D - if so, how might I enable it? Whilst the min-dist of 0.7m is by no means a showstopper for one of my applications, being able to reduce it would be incredibly useful.