Menu
What's new

Blue Iris and CodeProject.AI ALPR

C

Crest6010

Young grasshopper
Joined
Nov 20, 2019
Messages
38
Reaction score
5
Location
Maryland
Maryland unfortunately has many plates with small letters stacked on top of each other like the picture below. I can never get a correct reading when these plates come through. Has anybody else had success with similar plates? Is this something that should be added into the model training or is there something else I can try? 722133BB-A1D3-4E29-BBA1-82690A720515.jpeg
 
I

Il_Pres

Getting the hang of it
Joined
Nov 20, 2020
Messages
51
Reaction score
25
Location
Italy
Hi all,
I've implemented the LPR in my Blue Iris setup using SenseAI, which seems more maintainable and fully local, if compared to my previous setup, that was using a PY script to call Plate Recognizer ALPR Server. All is working correctly, but the SenseAI OCR is not as precise and reliable as the PR server. This is resulting in wrong plates number returned, with an high percentage, from the same cams used with my old script and that were working perfectly with PR OCR.

Is there anyway to improve the OCR module of SenseAI ?

Or should I go Back to Plate Recognizer APIs? BTW, I think I could be able to modify the SenseAI scripts to call PR rather than internal OCR, which I wouldn't do, for maintenance implications.

Thank you!
 
MikeLud1

MikeLud1

IPCT Contributor
Joined
Apr 5, 2017
Messages
2,206
Reaction score
4,247
Location
Brooklyn, NY
wpiman said:
Any chance this could be added to the arm64 versions of Codeproject? I am playing around with a Orange Pi 5 which is is maybe 80% faster than a RPI 4. I am trying to compare that to my old Xeon server running windows.

I also might run this on an M1 Mac which can run ARM docker files... That processor appears to really haul ass..

That said, all my benchmarking shows the oldschool xeon E3s running codeproject.ai on windows as being the fastest still?
Click to expand...
I just bought a Orange Pi 5 and was play with it. I have AI working outside of CodeProject.AI next is to make a module CodeProject.AI. I am not sure when I will have it done, but I am working on it

1680746264112.png
 
W

wpiman

Pulling my weight
Joined
Jul 16, 2018
Messages
332
Reaction score
246
Location
massachusetts
MikeLud1 said:
I just bought a Orange Pi 5 and was play with it. I have AI working outside of CodeProject.AI next is to make a module CodeProject.AI. I am not sure when I will have it done, but I am working on it

View attachment 159220
Click to expand...
Awesome!!! It looks like it is a good SBC which is better than an RPi4, but given the shortage prices is pretty competitive.

Let me know how it works out. Did you get an NVME board for it? Which size memory did you get?

There is some sort of TPU (Tensor Flow Processing Unit) on there. I am not sure if that can be utilized. I think CodeProject uses Tensor flow. Docker was natively installed on their ubuntu install- so I was able to get the docker ARM image working..
 
W

wpiman

Pulling my weight
Joined
Jul 16, 2018
Messages
332
Reaction score
246
Location
massachusetts
coral.ai

USB Accelerator | Coral

A USB accessory that brings accelerated ML inferencing to existing systems. Works with Windows, Mac, and Raspberry Pi or other Linux systems.
coral.ai coral.ai

I wonder if there Coral Accelerators would improve things. Codeproject.ai claims they support them.
 
N

nhs128

Young grasshopper
Joined
Jul 8, 2017
Messages
91
Reaction score
21
I am trying to run CP AI on a separate machine than by BI. I've got 2 main problems at the moment. 1) I do not show any plates whatsoever. I only see DayPlate and NightPlate. 2) If I try to read a plate image, it only works very slow on CPU, not on GPU. I always get 'No predictions returned.' I've read through this forum but so far I've not been able to find a good solution. I've read a bit about the CUDA version and downgrading but not sure if I need to do that, or how. If it is the GPU, I will just upgrade it. I wanted to run this on my BI machine but it doesn't have a GPU and is a slim pc. Appreciate the help in advance.


1681500183935.png
 
MikeLud1

MikeLud1

IPCT Contributor
Joined
Apr 5, 2017
Messages
2,206
Reaction score
4,247
Location
Brooklyn, NY
nhs128 said:
I am trying to run CP AI on a separate machine than by BI. I've got 2 main problems at the moment. 1) I do not show any plates whatsoever. I only see DayPlate and NightPlate. 2) If I try to read a plate image, it only works very slow on CPU, not on GPU. I always get 'No predictions returned.' I've read through this forum but so far I've not been able to find a good solution. I've read a bit about the CUDA version and downgrading but not sure if I need to do that, or how. If it is the GPU, I will just upgrade it. I wanted to run this on my BI machine but it doesn't have a GPU and is a slim pc. Appreciate the help in advance.


View attachment 159992
Click to expand...
Can you post the below screenshot
1681501371632.png
 
N

nhs128

Young grasshopper
Joined
Jul 8, 2017
Messages
91
Reaction score
21
MikeLud1 said:
It looks like there is an issue with the ALPR module, you can try uninstalling it then reinstall or wait till CPAI v2.1.0 comes out. Which might be today.
Click to expand...
I tried reinstalling but get the same result. I will wait for the next version and test once it is available.
 
MikeLud1

MikeLud1

IPCT Contributor
Joined
Apr 5, 2017
Messages
2,206
Reaction score
4,247
Location
Brooklyn, NY
nhs128 said:
I tried reinstalling but get the same result. I will wait for the next version and test once it is available.
Click to expand...
Once the new version comes out folow the below steps to install

1. Uninstall CP:AI
2. Delete the Codeproject Directory completely - "C:\Program Files\Codeproject"
3. Delete "C:\ProgramData\CodeProject"
4. Restart The Computer
5. Reinstall CodeProject along with the modules that you wish to install.
6. Restart Computer one more time
7. Run BI and CP:AI
 
Tron

Tron

n3wb
Joined
Aug 12, 2017
Messages
10
Reaction score
11
@MikeLud1 - I'm in Australia and using the license plate module with a high percentage of success however I am getting a few anomalies.. Are you interested in training the AI module further by having us common folk like myself sharing licence plate pictures that have these issues via PM or via this thread? Or is this not something you are keen on?

Cheers
 
You must log in or register to reply here.
Top