Blue Iris and CodeProject.AI ALPR

Hey @MikeLud1 Is there any way that I can train the licence-plate model to work a bit better with Australian plates? Although in saying that I've got no clue how but willing to learn.
I've just started running @algertc database system for ALPR which is brilliant way to view things in an organised fashion and whilst your model picks up most plates it seems to get a bit funny with some.. please note a large majority of the plates are correct and it does an amazing job and I am purely wanting to help potentially improve things if I can help at all. Here are some examples

First one, some plates in Australia (In the ACT) have a euro design.
Example picture of a standard European style numberplate

Standard European style numberplates must start with the letter 'A' then a space which contains the ACT crest and bluebell logo, then a combination of 6 letters and numbers following the standard ACT 'Y' series format.
Here is one below going past my camera:
2025-05-08_16-38-26.jpg

Unsure what happened with this and how it got TEET lol:
1746686472178.png

A few of the slim-line plates seem to get picked up with an "E" at the start like below I've noticed as well:
1746686527667.png
 
Anyone else experiencing memory leak with the ALPR module?
Yes and like you I tried to keep it under control using an automation, but I found a better way through BI. Under the AI settings in BI there is an option to restart CPAI every x number of minutes. I set mine to 720 (every 12 hours) and it works great.


Screen Shot 2025-05-08 at 6.52.33 AM.png
 
Last edited:
  • Like
Reactions: Skinny1
Yes and like you I tried to keep it under control using an automation, but I found a better way through BI. Under the AI settings in BI there is an option to restart CPAI every x number of minutes. I set mine to 720 (every 12 hours) and it works great.


View attachment 220618

Thanks. This will certainly help someone out there! I'm not using Windows for my CodeProject.AI so this won't work for me, unfortunately.
 
How do I change the rotation degrees, or for that matter, height, width, etc?

I'm not exactly sure what the official way to do that is, I'm on a RKNN SBC, so it's a little different if you are running windows.

/app/modules/ALPR-RKNN/modulesettings.json

Has all of the settings, so I'd imagine you'd edit this file and restart the module.
 
  • Like
Reactions: JNDATHP
Interesting troubleshooting:

TLDR:

I'm having a similar issue with a newly installed CPAI 2.9.5 and LPR and thought I was losing my mind.

A little more info that may help:

Installed CPAI 2.9.5 with LPR 3.3.4 and Object Detection (YOLOv5.NET) 1.14.0 using the same Windows installer exe on multiple different installs.

The older installs are working fine using the same exact versions. The 2 installs I did over the past few days are experiencing that same problem as JNDATHP

Using CPAI Explorer the Object Detection works fine. Using CPAI Explorer LPR always returns "No prediction returned".

Looking at the logs (with trace) it did find the DayPlate but there is an error with OpenCV
1747110670972.png

Here is a screenshot of 2 successful Object Detections and then 2 failed LPR
1747111457730.png

Is it possible one of the dependences is messed up from the source? Like OpenCV or PaddleOCR? Just guessing since I've successfully installed CPAI like 20 times in the past year (lots of testing).

I have uninstalled and reinstalled multiple times in the last 2 days with "Do not use download cache" on and off. I tried both "Check for updates" on and off. Nothing helped.

I've actually successfully installed these same versions a few times a few months so I was surprised this time was giving me such a hard time these past couple of days.