camera picture for regions of interests is tilted against reference-image

[Harold-3235464]

The Problem

I am new here, so excuse if problem already has been solved, but I didnt't find it here.
The camera-picture on which the regions of interest are applied, is tilted against the prevousliy taken reference-image. Therefor the reading fails.
I repeated the reference-image-procedure already several times and same with regions of interest, but every time the reading picture is tilted after that.
It would be great to hear, what I am doing wrong.
Pictures to explain the topic:

If you need more data to support, please let me know

Version

Firmware version: Release: v15.3.0 (Commit: 3fbff0a) Firmware build time: 22.07.2023 09:42 Web interface version: Release: v15.3.0 (Commit: 3fbff0a)

Logfile

.

Expected Behavior

the camera-picture used for reading should be "same" orientation/angle as reference- and learnt ROIs

Screenshots

screenshots are appended in problem

Additional Context

.

[caco3]

It is possible that the reference data did not get saved to the SD card properly.

First tip would be to save the reference image and marks once mor3e, but you say you did this already.
You should set the log to DEBUG for those steps and try to really save and even restart between each step.

Also you could try to use a different SD-Card or check it got inserted properly.

[Harold-3235464]

Hello thanks for your response :),
I checked the SD-Card. Reinserted it. Then I made a new reference image and positioned the digital and analog reading-frames by try and error to the faulty position (not aligned to the picture but to the, so the watermeter was reading. Yesterday 29.08, this has been working for a while, but then the tilt changed. All the time, the life-stream does not change the angle.
Today, 30.09 the original tilt is back againg - without having changed anytining and after that back to the "wrong" tilt.

log_2023-11-29.txt
log_2023-11-30.txt
data_2023-11-30.csv

watermeter_2023-11-30_06-48-40.zip

Hoping the data will help to find out why the tilt changes.
Harold

[tronauts]

I have the same problem.

[friedpa]

The alignment marks are not good. Just stick a black dot on the watermeter at the bottom and take the m3 on the top. Secondly correct the lens so that the picture is sharper (your alignment mark on the left side has way to less contrast)

[Harold-3235464]

Hello, thanks for your advice. I will try to follow your instructions. Instead of M22 in the frame, I will choose the m³ and the dot. The dot is (originally) chosen. I will keep you updated about the result. :) Harold Ebeling Mohnweg 4 91578 Leutershausen Am Do., 30. Nov. 2023 um 15:29 Uhr schrieb Paul Fried < ***@***.***>:

…

The alignment marks are not good. Just stick a black dot on the watermeter at the bottom and take the m3 on the top. Secondly correct the lens so that the picture is sharper (your alignment mark on the left side has way to less contrast) — Reply to this email directly, view it on GitHub <#2718 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AXBP3WSCLJSD6MM54ESCAYLYHCJ6BAVCNFSM6AAAAABAAVXUK2VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM3TOMJYGU2DQ> . You are receiving this because you authored the thread.Message ID: <jomjol/AI-on-the-edge-device/repo-discussions/2718/comments/7718548@ github.com>

[caco3]

Also make sure that:

• The reference marks are not too large (increased alignment time)
• The reference marks are as far apart as possible
• The reference marks do not cover any changing parts (digit, pointer)

[Harold-3235464]

Hello Paul, thanks for your advice. This really helped. Now the picture is stable. :) But may be you may also have an advice for my second problem. I have a meter where the digits already will show the new digital value, but the analog meters still haven't finished their circles. So it always comes to misreadings. I 've been fighting with the last digit, and thought, that I had solved it, but now the second last digital number is changing and I have again the problem. As you can see, it now calculates with 100 instead of 109. So last digit again is switching too early to 0, but second does not follow (in that case it should be 1 => 110). How do I have to setup the configuration (6.x nor 9.9 does help in analog/digital transition.)? [image: grafik.png] thanks in advance for your support Harold Ebeling Mohnweg 4 91578 Leutershausen Am Do., 30. Nov. 2023 um 15:29 Uhr schrieb Paul Fried < ***@***.***>:

…

The alignment marks are not good. Just stick a black dot on the watermeter at the bottom and take the m3 on the top. Secondly correct the lens so that the picture is sharper (your alignment mark on the left side has way to less contrast) — Reply to this email directly, view it on GitHub <#2718 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AXBP3WSCLJSD6MM54ESCAYLYHCJ6BAVCNFSM6AAAAABAAVXUK2VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM3TOMJYGU2DQ> . You are receiving this because you authored the thread.Message ID: <jomjol/AI-on-the-edge-device/repo-discussions/2718/comments/7718548@ github.com>

[friedpa]

Entschuldigung für meine späte Antwort, ich sehe leider das Bild nicht.....

[Harold-3235464]

Hallo Paul?, nochmals danke für die Unterstützung. Ich hatte das Bild direkt in die Mail kopiert. Jetzt ein erneuter Versuch mit einem aktuelleren Bild (gleiches Problem) als Anhang. In dem Bild siehst du einen Zähler 109.790. Ermittelt wird aber 110.79. LG Harold Ebeling Mohnweg 4 91578 Leutershausen Am Mi., 6. Dez. 2023 um 09:55 Uhr schrieb Paul Fried < ***@***.***>:

…

Entschuldigung für meine späte Antwort, ich sehe leider das Bild nicht..... — Reply to this email directly, view it on GitHub <#2718 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AXBP3WTEVJNGDASCFES57NTYIAXJBAVCNFSM6AAAAABAAVXUK2VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM3TONZTHAZDE> . You are receiving this because you authored the thread.Message ID: <jomjol/AI-on-the-edge-device/repo-discussions/2718/comments/7773822@ github.com>

[Marvel912]

I had same issues, even tried to upload an older version, updated region of interest and many more, but without success. finally get it working. i have adjusted both allignment marker to totally different position. Now it is working for me love it

[jasaw]

I have occasional alignment issue and it looks like the image is rotated a bit too much. Looking at the set up process in its entirety, the rotation should be limited to less than a couple of degrees (probably make it configurable).

The reason being:
The camera is mounted as the first step, then a reference image is taken. If the camera is moved, a new reference image has to be taken for things to work optimally anyway. Taking a new reference image requires new alignment markers to be taken. From this point on, the images that are taken shouldn't need a lot of rotation. If the algorithm thinks that it needs 30 degrees rotation, that's obviously wrong, so it makes sense to limit rotation and/or give up on rotation if it can't find the alignment markers.

[caco3]

a limiter might be useful, but ir also is not expected for the camera to move after setup!

[caco3]

If the algorithm thinks that it needs 30 degrees rotation, that's obviously wrong, so it makes sense to limit rotation and/or give up on rotation

How much is the rotation off compared to the alignment image?

[jasaw]

How much is the rotation off compared to the alignment image?

I didn't measure it, but I estimate maybe 5 to 10 degrees off, which is enough to throw it off.

If the expectation is the camera shall not move after setup, then it would be better to have a configuration option to turn off alignment altogether (probably by setting max rotation config to zero).

[Slider0007]

a configuration option to turn off

There is already an expert parameter Alignment Algorithm implemented to switch off alignment marker mapping algorithm. If you switch this parameter to off only inital rotation (defined in reference image page) will be applied for correction, but no additional pattern mapping with alignment marker.

Same topic: #2736 (reply in thread)

[jasaw]

Thank you @Slider0007 . I didn't know the auto alignment can be turned off. It's working better now that it's turned off.