Hello, I own a OAK-D Pro PoE with Wide FOV. I'm noticing a bad depth accuracy. Indeed, the camera measures 3 meters while the real distance is 3.5 meters. The camera is placed on the ceiling and frames a plain floor. 3.5 meters is the central vertical distance. I used both AlphaScaling 1 and 0 without any difference. Thanks in advance

Hi LcicC! The depth accuracy issue you're experiencing could be due to several factors. One of the key things to note is that the Wide Field of View (FOV) lenses on the OAK-D Pro PoE are fixed-focus. This means that the depth accuracy can vary depending on the distance and the specific lens used. For the wide FOV version, you can select either a 120° DFOV IMX378 (12MP, rolling shutter) or a 150° DFOV color OV9782 (1MP, global shutter) color camera. The OAK-D Pro PoE uses a technique called Active Stereo Vision (ASV), where a laser dot projector projects many small dots in front of the device to help with disparity matching. This is especially useful for low-visual-interest surfaces such as a wall or floor. However, note that the laser dot projector and flood LED are disabled by default for safety reasons, and you would need to intentionally enable them. The depth perception of the OAK-D Pro PoE has an ideal range of 40cm - 6m, with a minimum depth perception (MinZ) of ~20cm (400P or 800P, extended) or ~37cm (800P), and a maximum depth perception (MaxZ) of ~10 meters with a variance of 10%. In terms of depth accuracy, it's important to note that depth accuracy can oscillate with distances due to the nature of disparity matching. This is a common characteristic of stereo depth cameras. The depth accuracy of a median device would be below 2% absolute depth error for distances below 3m, below 4% absolute depth error for distances between 3m - 6m, and below 6% absolute depth error for distances between 6m - 8m. If you're still experiencing issues with depth accuracy, it might be worth checking the calibration of your device, as factory calibration is typically optimal. You could also consider using postprocessing filters to improve the depth accuracy. Sources: - [OAK-D-PRO-W-POE - CAMERA-MODULE-SPECIFICATIONS](https://docs.luxonis.com/projects/hardware/en/latest/pages/NG9097prow#oak-d-pro-w-poe) - [OAK-D-PRO-W - HOW-IT-WORKS](https://docs.luxonis.com/projects/hardware/en/latest/pages/DM9098prow#oak-d-pro-w) - [OAK-D-PRO-W - STEREO-DEPTH-PERCEPTION](https://docs.luxonis.com/projects Hope this helps! :) Best, Luxonis BOT

> @"LcicC"#p25431 The camera is placed on the ceiling and frames a plain floor. How does the depth map look? Is it uniform or very noisy? If the floor is too texture-less the stereo algorithm might have some problems. Perhaps the calibration is bad. Could you give me the MXID of the device, I can check factory calibration. If that is the case, i'd recommend you recalibrate the device. https://docs.luxonis.com/hardware/platform/depth/calibration/ Thanks, Jaka

Hello, This is the MxId: 1844301041EFE9F400 Thank you in advance

@"LcicC"#p25593 No logs are available on our side. I'd suggest you recalibrate your device since it seems it still has the old factory calibration. https://docs.luxonis.com/hardware/platform/depth/calibration/ Thanks, Jaka

When following such tutorial, I get the error "**Failed to detect markers on the image**" in the calibration step. Indeed the camera does not recognize the grid of squares as expected: see attached image. [upl-image-preview url=https://discuss.luxonis.com/assets/files/2024-09-09/1725894580-188861-calib.png] The same happens when I use a 28" screen to display the Charuco board, with no particular brightness condition (neither saturated or with direct light). Any idea what's going on?

Depth Accuracy

jakaskerl

LcicC
Sorry, I must have missed the post...

The aruco markers are not detected.
Which repo version are you using, which depthai version?
What command do you use to call the calibrate.py?

Thanks,
Jaka

LcicC

Hi jakaskerl

Repo version: the lastest from GitHub - luxonis/depthai: DepthAI Python API utilities, examples, and tutorials., I guess its v3.6.1

Depthai version: also the lastest, 2.28.0.0

Command to call calibrate: python3 calibrate.py -s 3.3 –board OAK-D-PRO-W-POE -nx 7 -ny 4

Charuco board: the 28” screen one as per link in the tutorial Calibration (luxonis.com)

I also tried with an older version of depthai (2.24), different cameras, and printing the charuco board instead of displaying it on the screen.

jakaskerl

LcicC
Your board is 13 by 7 not 7 by 4.. Also, it looks like the board is very "wavy" and is terrible if you want to calibrate for camera distortions. Make sure it is as flat as you can possibly make it.

Thanks,
Jaka

LcicC

Using 28” screen, and command python3 calibrate.py -s 3.3 –board OAK-D-POE-W-97 -nx 15 -ny 8, it looks alright but then the calibrate.py gets stuck right after the three “INTRINSIC CALIBRATIONS”.

jakaskerl

LcicC
Is the charuco board actually 15x8? I think a window should pop up to check the epipolar error.

Thanks,
Jaka

LcicC

Yes, the board is 15x8 (not the same of the first image). My bad, there is the pop window and then it gets stuck.

jakaskerl

LcicC
Close the window?

LcicC

Hi Jaka,

I have waited both with the window opened, and after having closed the window. It gets stuck.

I’m using command python3 calibrate.py -s 3.6 –board OAK-D-POE-W-97 -nx 15 -ny 8, with charuco board for 28” screen and depthai version 2.28.0.0.

getEepromData() gives product name OAK-D-W-POE-97 and board name NG9097.

jakaskerl

LcicC
Ok I will take a look. I'll let you know if I can reproduce the issue.

Thanks,
Jaka

jakaskerl

Hi @LcicC
I tested the script and got no issues.

which window is the one that the script gets stuck on?
try closing it with spacebar; apparently some force close of windows bricks python...
Can you provide any images of right before the script dies and right after?

Thanks,
Jaka

LcicC

Hi Jaka,

Thaanks! It turns out all I had to do was press the spacebar to continue (!). Wouldn't it be helpful to have a message saying so? With the video, I thought the person was just closing the window in the interface, and the tutorial doesn't mention it. The script itself says to press the spacebar to continue in the first step, but (maybe it's just me!) I didn’t think to do that after the step of taking the images. It’s not just after the first pop-up; you need to press the spacebar multiple times until the calibration is fully done. Also, I assume it's set up this way so you can review and validate each pop-up before continuing, so wouldn’t it be nice if the tutorial explained that process too—like what to look for during each step?

Thanks again.