CodeProject.ai with License Plate Reader module - HowTo?

[actran]

@Nunofya It's on BI settings with CodeProject.AI logo on the right

See your screenshot, it's underneath the "Face recognition" checkbox

 

[Vettester]

I also noticed you are using alpr as a custom model. Is there even an alpr custom model available? Try using license-plate.

Here's how mine is configured:

 

[MikeLud1]

Open for suggestions. Not showing any logging in CPAI on my Z12E LPR cam.

View attachment 170746View attachment 170748View attachment 170749View attachment 170750View attachment 170751View attachment 170752View attachment 170753View attachment 170754View attachment 170755View attachment 170756View attachment 170757View attachment 170758View attachment 170759

You need to checkoff ALPR for plates, all other setting are fine

 

[Nunofya]

@Nunofya It's on BI settings with CodeProject.AI logo on the right

See your screenshot, it's underneath the "Face recognition" checkbox

 

[Nunofya]

You need to checkoff ALPR for plates, all other setting are fine

View attachment 170780

do you mean have a "blank" box or have a 'check' in the box?

 

[actran]

@Vettester Using license-plate is old config advice.

If you read @MikeLud1 more recent config advice Blue Iris and CodeProject.AI ALPR , alpr for custom model is recommended.

Here is a more direct answer Blue Iris and CodeProject.AI ALPR

 

[MikeLud1]

do you mean have a "blank" box or have a 'check' in the box?

Have a check in the box, below are my settings

 

[Vettester]

If you read @MikeLud1 more recent config advice Blue Iris and CodeProject.AI ALPR , alpr for custom model is recommended.

Hmmm... not sure how I missed this. I'll check it out, thanks!

Edit: This didn't work on my CPU only system so I had to revert back to my original settings.

 

[Nunofya]

Should "Cuda true" be green? Don't have a dedicated gpu.

 

[actran]

Hmmm... not sure how I missed this. I'll check it out, thanks!

Edit: This didn't work on my CPU only system so I had to revert back to my original settings.

@Vettester What BI version are you using?

@MikeLud1 says that his alpr advice requires a recent, recent BI version. Blue Iris and CodeProject.AI ALPR

 

[Vettester]

What BI version are you using?

BI - 5.7.9.6, CPAI - 2.1.11

 

[actran]

BI - 5.7.9.6, CPAI - 2.1.11

That's pretty recent, can't argue with that.

@Vettester The only other thing I can think of: Did you uncheck "Only for confirmed vehicle alerts"?

Here is my config

 

[Vettester]

Did you uncheck "Only for confirmed vehicle alerts"?

That was it. It appears to be duplicating the license plate number in the alert image. Is there anyway to stop this?

 

[MikeLud1]

That was it. It appears to be duplicating the license plate number in the alert image. Is there anyway to stop this?

View attachment 170791

Not that I know of.

 

[Vettester]

Not that I know of.

Thanks Mike, I'll just have to learn to control my OCD.

 

[MikeLud1]

Thanks Mike, I'll just have to learn to control my OCD.

UI3 only shows the one plate

 

[actran]

@Vettester IMHO, they're not duplicates. What's displayed is %PLATE; %MEMO in BI5 console

 

[Vettester]

UI3 only shows the one plate

After watching this for a few hours I noticed that I'm missing a lot of plates that clearly should have been recognized. Switching back to the license-plate model seems significantly more reliable.

 

[MikeLud1]

After watching this for a few hours I noticed that I'm missing a lot of plates that clearly should have been recognized. Switching back to the license-plate model seems significantly more reliable.

View attachment 170800

View attachment 170801

It should not make a difference because the ALPR module uses the license-plate model to find the plate before it does the OCR. Post a screenshot of your camera's AI settings

            start_time = time.perf_counter()
            detect_plate_response = await module_runner.call_api("vision/custom/license-plate",
                                                                 files={ "image": image_data },
                                                                 data={"min_confidence": opts.plate_confidence})
            inferenceMs += int((time.perf_counter() - start_time) * 1000)

 

[MikeLud1]

After watching this for a few hours I noticed that I'm missing a lot of plates that clearly should have been recognized. Switching back to the license-plate model seems significantly more reliable.

View attachment 170800

View attachment 170801

Below are my settings