OAK-D Pro PoE AF Running issue

RyanLee

We are repeatedly seen the error below with OAK-D Pro PoE AF. Please check it out and help us if you have any solution.

Best regards,
Ryan.

erik

Hi RyanLee ,
I haven't heard of overheating issue on the Pro PoE yet, that's new. After how long does it happen? Does it happen on all devices, or only specific one(s)?
Thanks, Erik

RyanLee

Hi erik

We've tested one of the three, and we are going to test the rest soon. We will share its test result as soon as possible.

The error we reported is showing below three cases.

When running the DepthAI program
referenced site: https://docs.luxonis.com/en/latest/pages/tutorials/first_steps/#first-steps-with-depthai

When using AI-related functions
referenced site: https://github.com/luxonis/depthai-experiments/tree/master/gen2-yolo
When the resolution of the camera is increased
When using other resolutions except THE_1080_P, a runtime error related to temperature occurs.
referenced site: https://docs.luxonis.com/projects/hardware/en/latest/pages/guides/getting-started-with-poe.html?highlight=rtsp#video-streaming-with-oak ==> RTSP streaming ==> https://github.com/luxonis/depthai-experiments/tree/master/gen2-rtsp-streaming

Thanks,

Best regards,
Ryan.

erik

Hi RyanLee ,
We apologize for this issue, please contact our Ops team and we will send a replacement unit. We would love to also get the faulty OAK camera to our offices for further investigation, as I haven't heard about thermal issues on OAK POE models yet.

Regarding 1080P resolution - could you share further data? Perhaps debug mode enabled on different resolutions - what temperature does sensor reach in, say, 3min, 5min, 10min, 15min, 30min.. at different resolutions? I would be really interested to know those, as perhaps it's something with firmware as well, not a faulty HW (heat dissipation).

Thanks, Erik

RyanLee

Hi erik

We are considering two OAK-D Pro W(OV9782) cameras. And it will look up slightly at a height of about 90 to check the people and their face, and another one that looks at the floor for cliff and chairs/stools recognition at a height of about 20 cm.

Here we have two questions.
First,

As I explain above, we will put two OAK-D Pro W(OV9782) cameras. And one of that will be installed with around 20cm height from floor. Do you think with that height OAK-D Pro W(OV9782 has any expected issue coming?? We hope to cross-check about it if it is possible.

Second,
Are there any issue you have met using OAK-D Pro W(OV9782) with the maximum dot projector output as 1.2A? We hope to cross-check about it before we finalized our decision to put your solution on our patrol robot.

Best Regards,
Ryan.

erik

Hi RyanLee ,

I can't think of any issue immediately, except that just note the minimal depth perception. For OAK-D Pro W it's MinZ: ~20cm (400P OR 800P, extended), ~37cm (800P).
As long as you can provide enough power there shouldn't be an issue. Just note that in some cases (eg. complete darkness), dots would be very visible and really large, which might actually decrease depth map quality compared to having tiny dots all around (as that's better for disparity matching).

Thanks, Erik

erik

Also note: The strength of the dots (luminescence) is the highest at 765 mA (30% duty cycle), and decreases as the current increases up to 1200 mA (6% duty cycle). So having it to max (1.2A) isn't actually the brightest.

RyanLee

erik

As I shared the robot will install two OAK-D-Pro(OV9782). One is for face recognition and people detection which will be installed top side of the robot and the other one is for cliff detection which will be installed bottom side of the robot.
We have a question and cross-check about the bottom side OAK-D Pro(OV9782) for cliff detection.
It will be installed 20cm height or 15 height. We are trying to find best position for its function of cliff detection.

<First Question>
So we hope to cross-check about it once more.

First one: I am wondering about you comments with “it's MinZ: _20cm (400P OR 800P, extended), _37cm (800P)”. Is it (400P OR 800P, extended) which case? Where can I check its details. And what about this case _37cm (800P)?? Could you please give me some more explanation which sensor or cases for??
I think that I do not match your information to your specification below image.
Second One: Again from you comment “it's MinZ: _20cm (400P OR 800P, extended), _37cm (800P)”. Does “MinZ: _20cm” mean the information from your specification information like below??
Third One: We don’t want to make any Blind Spot for its cliff detection to protect our robot.
So how do you think installing its camera like below. We think that we can avoid its blind stop with this. And if we install its camera with 15cm height, only think we need to make more than “MinZ: _20cm”. is it the right understand from you comment in the previous answer?

Fourth One: Do you have any cliff detection example with AI?? Or please guide us what is good way to make the function of cliff detection.

<Second question>
And we also would like to cross-check about IR LED and dot projector. Our robot will be working not only day but also night without its light. So we would like to use IR LED and dot projector together. And as I told you our function is cliff detection and it is critical function for robot safety. In this case do you have any recommendation to setup its current for IR LED and dot projector? Please guide to make best performance with its hardware for our function of cliff detection If you have any.

erik

Hi RyanLee ,

Great questions🙂

1.1) Both in case of 400P or 800P+extended you will get MinZ of 20cm. That's because extended disparity mode basically downscales frame by 2 (so 800P -> 400P), runs stereo depth, then upscales the depth and combines it for final stereo depth result.
1.2) That is for camera focus, not stereo depth, which isn't exactly the same.
1.3) If you need to detect depth directly in front (at the base) of the robot, then yes your mounting would likely be the best. If 10cm blindspot would be acceptable, then you could have further perception. You could also mount the OAK camera vertically to increase the vertical FOV to about 100deg (and decrease horizontal FOV to 70deg).

2) What's the approach for cliff detection? If it's based on pointcloud analysis, then you need only depth, and you can just use the IR dot projector (no need for illumination LED).

Thoughts?
Thanks, Erik

RyanLee

erik Hi Erik,

Thank you for your quick response and about the feedback for "2)".
Our robot will be opearationg at night without light. So we would consider illumination LED.
And if it is possible, Could you recommand its supply current value for illumination LED and dot projector each for best performance with the function of cliff detection??

Best regards,
Ryan.

erik

Hi RyanLee ,
It depends based on the environment. In the dark environments and close distances, it's completely different than bright environments and long range. It might be possible to create a controller for this, I will have a chat with the team about it.
Thanks, Erik