16.0.0rc5 rotates regular pictures differently than reference picture

[Zugschlus]

The Problem

Mein Reference Picture ist richtig rotiert. Die ROIs sind richtig markiert. In "Overview" ist das Bild um locker 10 ° nach rechts gedreht, die ROIs passen nicht. Drehe ich das Bild in Settings => Reference Image zu weit nach links, geht das Bild in "Overview" auch nach links, aber die ROI-Koordinaten passen nicht. Das ist leider unbenutzbar.

Version

v16.0.0-RC5 (Commit: 7836323+)

Logfile

[0d00h00m00s] 2024-12-08T13:45:08 <INF> [MAIN] =================================================
[0d00h00m00s] 2024-12-08T13:45:08 <INF> [MAIN] ==================== Start ======================
[0d00h00m00s] 2024-12-08T13:45:08 <INF> [MAIN] =================================================
[0d00h00m00s] 2024-12-08T13:45:08 <INF> [MAIN] PSRAM size: 8388608 byte (8MB / 64MBit)
[0d00h00m00s] 2024-12-08T13:45:08 <INF> [MAIN] Total heap: 4382863 byte
[0d00h00m02s] 2024-12-08T13:45:10 <INF> [MAIN] Camera info: PID: 0x26, VER: 0x42, MIDL: 0x7f, MIDH: 0xa2
[0d00h00m02s] 2024-12-08T13:45:10 <INF> [SDCARD] Basic R/W check started...
[0d00h00m02s] 2024-12-08T13:45:10 <INF> [SDCARD] Basic R/W check successful
[0d00h00m02s] 2024-12-08T13:45:10 <INF> [SNTP] TimeServer: ntp.ka51.zugschlus.de
[0d00h00m03s] 2024-12-08T13:45:10 <INF> [SNTP] Configuring NTP Client...
[0d00h00m03s] 2024-12-08T14:45:10 <INF> [SNTP] Time zone set to CET-1CEST,M3.5.0,M10.5.0/3
[0d00h00m03s] 2024-12-08T14:45:10 <INF> [SNTP] time zone: +0100 Delta to UTC: 3600 seconds
[0d00h00m03s] 2024-12-08T14:45:10 <INF> [SNTP] Time is already set: 2024-12-08 14:45:10
[0d00h00m03s] 2024-12-08T14:45:11 <INF> [MAIN] CPU frequency: 160 MHz
[0d00h00m03s] 2024-12-08T14:45:11 <INF> [SDCARD] Folder/file presence check started...
[0d00h00m03s] 2024-12-08T14:45:11 <INF> [SDCARD] Folder/file presence check successful
[0d00h00m03s] 2024-12-08T14:45:11 <INF> [MAIN] Tag: 'v16.0.0-RC5', Release: v16.0.0-RC5 (Commit: 7836323+), Date/Time: 2024-12-05 00:08, Web UI: Release: v16.0.0-RC5 (Commit: 7836323+)
[0d00h00m03s] 2024-12-08T14:45:11 <INF> [MAIN] Reset reason: Via esp_restart
[0d00h00m03s] 2024-12-08T14:45:11 <INF> [WLANINI] SSID: incluesionPPSK
[0d00h00m03s] 2024-12-08T14:45:11 <INF> [WLANINI] Password: XXXXXXXX
[0d00h00m03s] 2024-12-08T14:45:11 <INF> [WLANINI] Hostname: esp-watermeter
[0d00h00m03s] 2024-12-08T14:45:11 <INF> [WLANINI] RSSIThreshold: 0
[0d00h00m04s] 2024-12-08T14:45:11 <INF> [MAIN] WLAN config loaded, init WIFI...
[0d00h00m04s] 2024-12-08T14:45:11 <INF> [WIFI] Automatic interface config --> Use DHCP service
[0d00h00m04s] 2024-12-08T14:45:12 <INF> [WIFI] Set hostname to: esp-watermeter
[0d00h00m04s] 2024-12-08T14:45:12 <INF> [WIFI] Init successful
[0d00h00m06s] 2024-12-08T14:45:14 <INF> [WIFI] Connected to: incluesionPPSK, RSSI: -65
[0d00h00m08s] 2024-12-08T14:45:16 <INF> [WIFI] Assigned IP: 192.168.196.86
[0d00h00m24s] 2024-12-08T14:45:32 <INF> [MAIN] Device info: CPU cores: 2, Chip revision: 100
[0d00h00m24s] 2024-12-08T14:45:32 <INF> [MAIN] SD card info: Name: ASTC, Capacity: 7454MB, Free: 7425MB
[0d00h00m26s] 2024-12-08T14:45:34 <INF> [MAIN] Initialization completed successfully
[0d00h00m28s] 2024-12-08T14:45:36 <INF> [TFLITE] Trying to load the model. If it crashes here, it ist most likely due to a corrupted model!
[0d00h00m29s] 2024-12-08T14:45:37 <INF> [TFLITE] Trying to load the model. If it crashes here, it ist most likely due to a corrupted model!
[0d00h00m29s] 2024-12-08T14:45:37 <INF> [LOGFILE] Set log level to ERROR
[0d00h01m22s] 2024-12-08T14:46:30 <ERR> [POSTPROC] name: Raw: 35004.05515, Value: , Status: Rate too high - Read: 35004.05515 - Pre: 491.06646 - Rate: 34512.98869
[0d00h04m22s] 2024-12-08T14:49:29 <ERR> [POSTPROC] name: Raw: 35004.05515, Value: , Status: Rate too high - Read: 35004.05515 - Pre: 491.06646 - Rate: 34512.98869

Expected Behavior

Bild richtig gedreht, Erkennung Okay.

Screenshots

No response

Additional Context

No response

[Zugschlus]

I tried going back to the rolling release I was running before, but that doesn't get out of the Initialization phase any more. So upgraded to 16.0.0RC5 again, and now again with this bug here.

[SybexX]

This will be due to your image quality or your alignment marks. Improve the image quality or choose something other than alignment marks and don't make them too big.
Without reasonable/sufficient information from you, we cannot help you properly.

[Zugschlus]

Does the rotation of the image nowadays depend on the alignment marks? I always thought that the image is first rotated and then the alignment marks get applied?

[Zugschlus]

There is also something wrong with saving the alignment markers. See one screenshot after adjusting the marker. I then go back to Overview, Call up Settings -> Alignment Markers again, and the marker is wrong again and not what I set before.

[SybexX]

yes, the rotation is applied first and then the alignment marks are used for fine adjustment.
Aligning by alignment marks includes up, down, left right and rotation.
I just tried saving new alignment marks and couldn't find any errors.
Did you only select firmware.bin when updating or AI-on-the-edge-device__update__v16.0.0-RC5.zip?

[Zugschlus]

I updated using the AI on the edge update zip. I think I might have not pressed update / save / update / save in the correct order, after retrying a couple of times with different orders of button presses things are fine now. This should not have happened in the first place, but it might be a hard to diagnose usability issue which I understand you wouldn't want to address at the RC5 stage. Feel free to close.

[SybexX]

[SybexX]

@Zugschlus If you want to restore an earlier version, you must use the appropriate config.ini, as the migration function from the previous version does not migrate the config.ini back.

And now to this problem: what do you expect from us now or what solutions do you have to improve this?

[Zugschlus]

That would be kind of a usability issue. I have been using aiontheedge for years now and this is the first time I stumbled upon this. Is still think that the picture taken during a regular run being rotated grossly different from a newly taken reference picture might be unrelated.

Anyway, since I cannot reproduce this any more, there is probably nothing more to do than (1) thinking about usability of the reference image configuration and (2) keeping in mind that I experienced the rotation issue when the next user turns up complaining that their pictures taken during the normal run are off what the reference screen says.

Sorry for the noise.

[SybexX]

@Zugschlus to 2)
This has happened from time to time and it was always due to poor image quality or poorly placed alignment marks.
If everyone follow the instructions "correctly", this problem should not occur.
The reference image is only used to align/create the alignment marks and the ROIs; it is not needed or processed during the ongoing process. It is therefore important to set new alignment marks and ROIs whenever the reference image is changed.

• change camera settings >>> create new reference image >>> set alignment marks >>> set ROIs >>> reboot
• create new reference image >>> set alignment marks >>> set ROIs >>> reboot
• set alignment marks >>> reboot
• set ROIs >>> reboot

[Zugschlus]

I have now noted the instructions. Thankfully, I do not need to understand the connection between "rotate 79 degrees right" and wrongly set alignment marks while a change in the settings results in the image taken during the normal rounds changing its rotation as well. So I can just shrug it away while shaking my head.

Please feel free to close this issue.

[SybexX]

The rotation serves to accelerate the alignment process, during which the set rotation value is also exceeded/undershot. This provides additional image alignment if the camera changes position for any reason. Therefore, if you have a permanently installed camera, you can also switch off the alignment algorithm to shorten the processing time.