On-premise OCR Solutions for LPR?

C

cjowers

Getting the hang of it
Jan 28, 2020
107
36
AUS
Hi All,

Has anyone tried a home-made solution to read plates? For instance, plate recognition (and image crop) by YOLOv3, then OCR by pytesseract?

It seems there are plenty of examples out there, but the open source ones I've been looking at seem to be pretty bad at the character recognition side of things (why??) despite the cropping/thresholding/contour result giving a very good looking plate (which would be 100% obvious to a person). Is pytesseract still the best OCR we have? Or does the plate character recognition need to be trained for the specific font on your license plates for it to work reliably (with pytesseract i notice errors with O vs U, 4 vs G, etc.) ? Are there better methods (free/open source/on premise) for OCR of plates? I'd like to play around with this, so am looking for a decent OCR method, without going too overboard. Any thoughts?

Note: I'm starting from scratch, so open to ideas and just wanting to check in before I begin installing things. I'm not really interested in the free-trial cloud API keys some projects offer. This is mostly as a learning exercise, but it would be nice if it was worth doing. I think the LP detection and cropping would at least be useful to me, even if the OCR is no good.

Thanks


some examples that turned up in my brief search:
License Plate Recognition with OpenCV and Tesseract OCR - GeeksforGeeks
sergiomsilva/alpr-unconstrained
alitourani/yolo-license-plate-detection
TheophileBuy/LicensePlateRecognition
 
  • Like
Reactions: djernie
what is your OS? linux or Windoze?
Click to expand...
windows (10 pro)
(have docker installed if it helps)

or if there are better solutions for linux, im interested to hear. I could probably setup a dual boot PC for development purposes.
 
Last edited:
cjowers said:
windows (10 pro)
(have docker installed if it helps)

or if there are better solutions for linux, im interested to hear. I could probably setup a dual boot PC for development purposes.
Click to expand...

I have setup a web based ocr, the client uses curl to upload file.
Scripts are for linux
download from Optasia Download
pls read the readme therein
 
Thanks for this! I'll look into it. As I was looking up Optasia, one of your videos popped up as well. It looks really good, pretty cool that you only need the single board computer. Is that model similar in specs to a RasPi 4?

Optasia looks really good for multi-cam setups on public roads, with the parallel, relational sql-like queries for various checks (speed and traffic laws, amber alerts, etc) and probabilistic predicates to ignore most things.
Just so I understand, did you create your own web-based ocr, or are you using an exisiting one to return results?

Thanks again, really neat stuff
 
cjowers said:
Optasia looks really good for multi-cam setups on public roads, with the parallel, relational sql-like queries for various checks (speed and traffic laws, amber alerts, etc) and probabilistic predicates to ignore most things.
Click to expand...
we are not related to this article
our videos are here

Just so I understand, did you create your own web-based ocr, or are you using an exisiting one to return results?

Thanks again, really neat stuff
Click to expand...

the ocr is our own.
 
very cool, I'll check it out and start with the readme. And any other good resources you can recommend for learning this space would be appreciated too or general tips on this setup.
Thanks

-just found some more articles and videos on your system. the exposure bracketing with gige/usb3 cameras looks like it handles variability well. Does it wait for a suspected plate before it changes exposure, or does it just loop through exposures continuously? is this only for cams that don't need a continuous live feed (and not IP cams of course)?
www.vision-systems.com

Neural networks help identify license plates for traffic control

Combining off-the-shelf cameras to a PC running neural network software allows Singapore authorities to perform traffic monitoring and enforcement in an intelligent transportation...
www.vision-systems.com www.vision-systems.com
 
Last edited:
samplenhold said:
So does your company (Survision) provide software similar to other vendors that home based LPR users can use at a reasonable price?
Click to expand...
Our company doesn't provide software. The camera is exclusively made for LPR, there is a firmware embedded inside the camera that will read the license plate and send the data to whatever system, or simply by Wiegand.
 
cjowers said:
Hi All,

Has anyone tried a home-made solution to read plates? For instance, plate recognition (and image crop) by YOLOv3, then OCR by pytesseract?

It seems there are plenty of examples out there, but the open source ones I've been looking at seem to be pretty bad at the character recognition side of things (why??) despite the cropping/thresholding/contour result giving a very good looking plate (which would be 100% obvious to a person). Is pytesseract still the best OCR we have? Or does the plate character recognition need to be trained for the specific font on your license plates for it to work reliably (with pytesseract i notice errors with O vs U, 4 vs G, etc.) ? Are there better methods (free/open source/on premise) for OCR of plates? I'd like to play around with this, so am looking for a decent OCR method, without going too overboard. Any thoughts?

Note: I'm starting from scratch, so open to ideas and just wanting to check in before I begin installing things. I'm not really interested in the free-trial cloud API keys some projects offer. This is mostly as a learning exercise, but it would be nice if it was worth doing. I think the LP detection and cropping would at least be useful to me, even if the OCR is no good.

Thanks


some examples that turned up in my brief search:
License Plate Recognition with OpenCV and Tesseract OCR - GeeksforGeeks
sergiomsilva/alpr-unconstrained
alitourani/yolo-license-plate-detection
TheophileBuy/LicensePlateRecognition
Click to expand...

cjowers said:
Hi All,

Has anyone tried a home-made solution to read plates? For instance, plate recognition (and image crop) by YOLOv3, then OCR by pytesseract?

It seems there are plenty of examples out there, but the open source ones I've been looking at seem to be pretty bad at the character recognition side of things (why??) despite the cropping/thresholding/contour result giving a very good looking plate (which would be 100% obvious to a person). Is pytesseract still the best OCR we have? Or does the plate character recognition need to be trained for the specific font on your license plates for it to work reliably (with pytesseract i notice errors with O vs U, 4 vs G, etc.) ? Are there better methods (free/open source/on premise) for OCR of plates? I'd like to play around with this, so am looking for a decent OCR method, without going too overboard. Any thoughts?

Note: I'm starting from scratch, so open to ideas and just wanting to check in before I begin installing things. I'm not really interested in the free-trial cloud API keys some projects offer. This is mostly as a learning exercise, but it would be nice if it was worth doing. I think the LP detection and cropping would at least be useful to me, even if the OCR is no good.

Thanks


some examples that turned up in my brief search:
License Plate Recognition with OpenCV and Tesseract OCR - GeeksforGeeks
sergiomsilva/alpr-unconstrained
alitourani/yolo-license-plate-detection
TheophileBuy/LicensePlateRecognition
Click to expand...

Well, the accuracy of OCR can be impacted by various factors such as image resolution, contrast, and noise.
 
cjowers said:
Hi All,

Has anyone tried a home-made solution to read plates? For instance, plate recognition (and image crop) by YOLOv3, then OCR by pytesseract?

It seems there are plenty of examples out there, but the open source ones I've been looking at seem to be pretty bad at the character recognition side of things (why??) despite the cropping/thresholding/contour result giving a very good looking plate (which would be 100% obvious to a person). Is pytesseract still the best OCR we have? Or does the plate character recognition need to be trained for the specific font on your license plates for it to work reliably (with pytesseract i notice errors with O vs U, 4 vs G, etc.) ? Are there better methods (free/open source/on premise) for OCR of plates? I'd like to play around with this, so am looking for a decent OCR method, without going too overboard. Any thoughts?

Note: I'm starting from scratch, so open to ideas and just wanting to check in before I begin installing things. I'm not really interested in the free-trial cloud API keys some projects offer. This is mostly as a learning exercise, but it would be nice if it was worth doing. I think the LP detection and cropping would at least be useful to me, even if the OCR is no good.

Thanks


some examples that turned up in my brief search:
License Plate Recognition with OpenCV and Tesseract OCR - GeeksforGeeks
sergiomsilva/alpr-unconstrained
alitourani/yolo-license-plate-detection
TheophileBuy/LicensePlateRecognition
Click to expand...

Yes, it's possible to build a license plate recognition system using a combination of computer vision techniques and OCR. But the accuracy of OCR can be affected by the quality of the input image and the font used. To improve it, you may pre-process the input image. This can include techniques such as thresholding, filtering, and morphological operations. However, depending on the specific task, there may be other OCR software that may be more suitable. So it would be helpful to seek professional advice from experts, such as those from Document Scanning and OCR SDK for Apps & Web – Smart Engines. This can help to reduce errors and improve recognition accuracy.
 
Last edited:
^^ last wo posts: @EvieHoughton & @HolmesRebecca :

You do realize that this thread has been stale for almost two years, right?

Are both of you the same person? You are both 32 and joined this morning and this is your one and only post? Made this post and then skedaddled?
 
  • Like
Reactions: looney2ns
You must log in or register to reply here.
Top Bottom