Oak-D pro depth accuracy specifications

Tsjarly

Hello,

In the docs you specify the depth accuracy for the Oak-d pro camera with an error around 1% at a distance between 1 and 2 meters. You also mention that the ideal depth range starts from 70cm onwards. And with using extended disparity, the working range starts at 35cm already.

Could you also include this range of operation (0.35 - 1m) in your depth accuracy specification, as our application is mostly between 35cm and 1.5m

jakaskerl

Hi @Tsjarly
Not currently possible with the way we measure depth accuracy. To achieve (expected) sub % accuracy, we would need sub 3.5mm accuracy of the measurement system which we do not have atm.
We can extrapolate the values if we assume linear dependency and we should get a relatively good approximation of the error. All I can say with certainty is that the error is below 1%.

Thanks,
Jaka

Tsjarly

okay, just wondering whether the extrapolation would be fair, as in the graph (https://docs.luxonis.com/projects/hardware/en/latest/pages/guides/depth_accuracy/#p-75mm-baseline-distance-oaks) it looks like the behavior between 1 and 2 meters is already slightly different than for longer ranges, where it seems to move slightly upwards again, closer to the 1m point, so based on this graph I wouldnt exclude an error getting higher again for lower ranges. How can you be certain of an error below 1% if you cannot measure it?

Thanks,
Jarl

jakaskerl

Hi @Tsjarly

It's more accurate since there is more reference points for depth calculation. As you can see on the graphs, the depth gets better with closer distance, up to a point where either the stereo baseline is too high to allow overlap between the two cameras, OR the rail we use for testing is not accurate enough.

Thanks,
Jaka

Tsjarly

Hi @jakaskerl,

Thank you for the quick response with the graphs!
You mention that depth gets better with a shorter distance, but in the graph (the bottom one with the blue line), it seems as if the z-error gets higher again, moving from the 1m mark, closer to the 35cm mark, to an error of 2.5% at the closest range.
Or are you implying that this increase in error at lower distances is purely because of the inaccuracy in the test setup? It makes sense that the depth error gets relatively higher with a constant error, so I don't really understand the first two graphs, what do they represent in terms of the test setup?

best,
Jarl

jakaskerl

Tsjarly Or are you implying that this increase in error at lower distances is purely because of the inaccuracy in the test setup?

It's mostly due to inaccuracies in the test setup. As the device gets closer, the error increases because the inaccuracies in ground truth measurements play a much bigger role in RMSE calculation.

Thanks,
Jaka