New CodeProject.AI License Plate Recognition (YOLO11) Module
- Thread starter MikeLud1
- Start date
So I have your original ALPR detection running on a Linux server on this camera I call LPR. That works great. I point to that here... It works great.
The camera LPROZONE points to the local host which have YOLO 11 for license plates reading. I can read the plates just fine in the explorer, but at some point Blue Iris isn't getting them. I am using this CPAI server for a lot of other stuff and that seems ok.
(Is the server having issues? Posting pictures is hit or miss. I'll attach the pictures.)
The camera LPROZONE points to the local host which have YOLO 11 for license plates reading. I can read the plates just fine in the explorer, but at some point Blue Iris isn't getting them. I am using this CPAI server for a lot of other stuff and that seems ok.
(Is the server having issues? Posting pictures is hit or miss. I'll attach the pictures.)
Not sure what is wrong. I can submit some photos and not others??!! Here is the LPR-- old one. Points to my Linux Server. On the clone I point to the local host (unclick override)... Turned off the addition models. If I leave that on, Yolo 5 6.5 complains.
I tried that... I think I got it working again.. Make a couple of local changes...
modulesettings.json (two fixes):
- Module ID key renamed ALPRYOLO11 → ALPR — Blue Iris seems only recognizes the standard ALPR module ID
- Route command renamed detect → alpr — Blue Iris validates the command field in the response matches alpr
It was weird-- it worked for a bit. I think since I clone the camera maybe BI gets confused.
..
modulesettings.json (two fixes):
- Module ID key renamed ALPRYOLO11 → ALPR — Blue Iris seems only recognizes the standard ALPR module ID
- Route command renamed detect → alpr — Blue Iris validates the command field in the response matches alpr
It was weird-- it worked for a bit. I think since I clone the camera maybe BI gets confused.
..
So those changes I made worked for a bit.... and then went to crap. I have 3 clones of the camera--
One -> original ALPR on my Linux NAS-- works great.. but slow on my Linux CPU box.
Two -> Mike Lud Yolo 11 ALPR-- windows/CUDA... worked on blue IRIS and then broke. Works fine on the server debug.
Three -> My bastardized version of your ALPR 11 to work on a Jetson Nano... Worked and then broke... Works fine of the server debug.
One thing I noticed was I turned on Mesh just to see the models, and my Linux had "image/alpr" while the other two had "vision/alpr"...
On my Jetson, I added a SECOND route to modulesettings.json and that seemed to fix things?!?!
{
"Name": "ALPR Detection (image)",
"Route": "image/alpr",
"Method": "POST",
"Command": "alpr",
...
}
I did that on the Yolo 11 ALPR on my local codeproject and that worked for me there too. Wondering if the fact I clone cameras causes blue iris some confusion.
One -> original ALPR on my Linux NAS-- works great.. but slow on my Linux CPU box.
Two -> Mike Lud Yolo 11 ALPR-- windows/CUDA... worked on blue IRIS and then broke. Works fine on the server debug.
Three -> My bastardized version of your ALPR 11 to work on a Jetson Nano... Worked and then broke... Works fine of the server debug.
One thing I noticed was I turned on Mesh just to see the models, and my Linux had "image/alpr" while the other two had "vision/alpr"...
On my Jetson, I added a SECOND route to modulesettings.json and that seemed to fix things?!?!
{
"Name": "ALPR Detection (image)",
"Route": "image/alpr",
"Method": "POST",
"Command": "alpr",
...
}
I did that on the Yolo 11 ALPR on my local codeproject and that worked for me there too. Wondering if the fact I clone cameras causes blue iris some confusion.
I'm not using CPAI, but I do have 2 cloned LPR cameras in BI and I'm not having any issues.
iwanttosee
Getting comfortable
I don't know how to say this other than "kind of". I took the ONNX file his directory and got it running in docker. What is your setup?
iwanttosee
Getting comfortable
The Docker file I built starts with a Jetson Image which is ARM only. I tried to start it on my x86 Linux and it died a quick death.
So you don't have a CUDA GPU on that system?
I actually have a a x86 with Linux running Docker without a GPU and run the old one. It runs about the same it terms of accuracy IMHO, though I have not run any tests. What are you looking for specifically? The speed?
iwanttosee
Getting comfortable
No, I don't use CUDA GPU. Tried a P620 with 512 CUDA cores in the past, it didn't help lower the detection number and the only things changed were the wattage used by system and the heat it created.
I'm currently using YOLO5 .net, so I wanted to see what YOLO11 has to offer.
Thanks for your help.