FOV Discrepancy OAK-D Pro W (IMX378)

TonyDay

Hello All,

I am currently integrating an OAK-D camera into a product and am having trouble getting the measured FOV to match with the quoted ones in the documentation.

I will use depth as an example but I am finding a similar thing with the RGB image.

So, we have 2xOV9282 with a H/V FOV: 127°/79.5° and a projector with H/V FOI 78°/61° (± 7% on both). The thing that I am assuming is limiting the depth FOV is the lower of the numbers, i.e. the FOI of the projector. However, when I make measurements of what is being detected I get a H/V FOV of 46° / 52°, which is way lower than the projectors FOI at the bottom of its tolerance.

FOI stated at 50% which I assume is intensity?

Can anyone shed some light (deliberate pun) on the situation?

Tony

jakaskerl

Hi @TonyDay
The projector doesn't have an effect on the FOV of depth image. You can see that by how the pattern doesn't cover the full FOV of the camera.
What does affect it is the rectification and undistortion process which by nature lowers FOV because not all "pixels" seen by one sensor are also seen by the other (therefore no stereo matching can occur).

Thanks,
Jaka

TonyDay

Hey Jaka,

Thanks for responding. Ok, so the IR projector doesn't have an affect meaning the FOV angle should now be the same as the individual sensors but offset in front of the camera, as shown in the images here. I understand that if one sensor is unable to see what the other can it cannot give depth information, but if presenting a flat surface to the camera I still get the FOV numbers I mentioned above. So is the dramatic drop in FOV caused purely by distortion?

Are the numbers I measured typical for this camera? or does it seem like something is wrong?

jakaskerl

Hi TonyDay
You can check the amount of undistortion by changing the alpha value of the stereo node (stereo.setAlphaScaling(0)). I think the drop in FOV is too large, but keep in mind the more wide angle the lens is, the more distortion, and the more pronounced the correction.

Maybe adding some images or MRE would help.

Thanks
Jaka

TonyDay

Hi jakaskerl

I must hang my head in shame, I have fallen victim to a basic mathematical error of scaling by 2. The FOV for the depth I am receiving is actually 80° H, 54° V. Does this sound like what the OAK-D Pro W should achieve? Matches nicely with the projector now, coincidently.

I assume there is a way to export the configuration, then I can share that with you?

Thanks

jakaskerl

Hi @TonyDay
Should be around 97deg.
Did you calculate the values or did you measure them? I think we calculated them so there might be some discrepancies.

Thanks,
Jaka

TonyDay

hi jakaskerl

This was from measured data, however I think the target may have been a bit uniform. I suspect the best thing to test with would be a QR code / AprilTag so the cameras can match features?

Kind regards,
Tony

fkmjec

jakaskerl
Hello Jaka!

I am currently confused; where did you get the 97deg number from in the specification? I did not find it anywhere, although I am getting it as the actual value for the FOV of the camera when I call self.calibData.getFov(dai.CameraBoardSocket(depthData.getInstanceNum()).

Thanks,
František

jakaskerl

Hi @TonyDay
I'll check if perhaps the calculations are wrong.

jakaskerl

fkmjec
It was on the old documentation, not sure why we didn't port it over. The specs for rectified depth FOV were:

`106° / 97° / 70° D/H/V

When getting the FOV, use useSpecTranslation=False so it is taken from calibration file not predifined fields.

Thanks,
Jaka

fkmjec

Ok, now I am next-level confused. Does this mean that the tech specs are just a lie? Or how does this reduction in FOV come to be?

jakaskerl

fkmjec
First the undistortion process removes some FOV - described here.

Then depth FOV is also lower because of perspectives. Green line is the FOV of depth.

Thanks,
Jaka