License Plate Recognition Using YOLOv4 Object Detection, OpenCV, and Tesseract OCR
ฝัง
- เผยแพร่เมื่อ 1 ต.ค. 2024
- Learn how to implement your very own license plate recognition using a custom YOLOv4 Object Detector, OpenCV, and Tesseract OCR! In this tutorial I will walk-through custom code I have created to run object detections to find license plates, crop the license plate region, preprocess the license plate using OpenCV, and then run it through Tesseract OCR to output the license plate number. The backbone of this code is TensorFlow and it is written in Python.
#objectdetection #OCR #licenseplaterecognition
This video covers the implementation of a custom YOLOv4 model to detect license plates and numerous preprocessing techniques such as blurring, thresholding, dilation, finding contours, and character segmentation. All of these are done in order to properly prepare the license plate in order for Tesseract OCR to extract the license plate text.
GET THE CODE HERE: github.com/the...
Video Breakdown:
1. Cloning or Downloading the Code for the Tutorial
2. Installing Dependencies and saving the custom YOLOv4 model
3. Explanation of the license plate recognition and code.
4. Running the License Plate Recognition on several images.
-------------------------Resources--------------------------
Train Your Own YOLOv4 Custom License Plate Object Detector in the Cloud: • YOLOv4 in the CLOUD: B...
Learn about and Run YOLOv4 Cropping Custom Function: • Crop and Save YOLOv4 O...
Running Object Counting Custom Function: • Counting Objects Using...
Running Pre-trained YOLOv4 model with TensorFlow, TFLite, TensorRT: • YOLOv4 Object Detectio...
running Custom YOLOv4 License Plate Detector Model with TensorFlow, TFLite, TensorRT: • How to Build a Custom ...
The Official YOLOv4 paper: arxiv.org/abs/...
If you enjoyed the video, please toss it a like! 👍
To Subscribe: / @theaiguy
Thanks so much for watching!
- The AI Guy
![[UNCUT] “บอสณวัฒน์” ลากไส้!! ใครหนุนหลัง "แม่ตั๊ก ป๋าเบียร์" I คนดังนั่งเคลียร์ I 30 ก.ย. 67](http://i.ytimg.com/vi/QlD7mU2KKRk/mqdefault.jpg)
Hope you enjoy this video! Let me know which field of AI or Computer Vision you want me to make videos on next? :)
Hi!) Always looking forward to your new videos !!! Thank you very much for what you are doing. Can you record a video of how to run several different models at the same time on one GPU in real time?
Егор Романов appreciate it! I will look into for sure! Thanks.
@@TheAIGuy You are so cool! Make yourself a sponosor button on the channel!
Maybe some satellite image analysis?
Continue with your great work! You are demonstrating that AI and CV is not so complicated as people think.
There are a lot of cool things that you could "dive into", such as, path following (focused on autonomous driving, it is such an amazing field and a lots of CV is used) and a real-time entrance system with facial recognition algorithms.
Hope you continue with your positive vibe and teaching in a such simple way. In college, my professors always advised myself to follow the KiSS protocol (Keep it Simple and Stupid) xD Thanks a lot for your work ;)
Hey man , thanks for the great work. One thing I am facing an issue with this code.
The command `python detect.py --weights ./checkpoints/ocr-416 --size 416 --model yolov4 --images ./data/images/car2.jpg --plate` doesn't throw any errors but it doesn't show the license plate number. My model is called `ocr-416`.
`License Plate #: ` is the only output i get.
Could you tell me what could be an issue and possibly how to fix it. Thanks ans keep up the great work
I fixed it. Gotta add tesseract to the path.
Add following line in utils.py :
pytesseract.pytesseract.tesseract_cmd = r'C:\Program Files\Tesseract-OCR\tesseract'
Add your installation path of tesseracr
@@krysis2155 hello I am facing the same issue can you please help
Thank you for tutorial! However, I am facing a little difficulty. I have installed the tesseract OCR clicking 'For Windows I recommend: Windows Install'
When I tried the command 'python detect.py --weights ./checkpoints/custom-416 --size 416 --model yolov4 --images ./data/images/car2.jpg --plate' the windows photo viewer opens and says says the image is removed or moved to another place but when I open the detection image from detection folder it is there with detected car and license plate also there is no number plate written over the bounding box and on terminal window it is also not displaying it.
I am facing the same problem
@@shehryar99 same as me, did anyone manage to solve it?
same for me. Any ideas how I can fix it please?
thank you for thias video but what's error here
> Overload resolution failed:
> - Can't parse 'pt2'. Sequence item with index 0 has a wrong type
> - Can't parse 'pt2'. Sequence item with index 0 has a wrong type
> - Can't parse 'rec'. Expected sequence length 4, got 2
> - Can't parse 'rec'. Expected sequence length 4, got 2
The video is very clear & crisp
Thank you so much for such content
Im facing getting the following error
License Plate #:
Traceback (most recent call last):
File "/content/drive/MyDrive/TheAIGuy/yolov4-custom-functions-master/detect.py", line 146, in
app.run(main)
File "/usr/local/lib/python3.10/dist-packages/absl/app.py", line 308, in run
_run_main(main, args)
File "/usr/local/lib/python3.10/dist-packages/absl/app.py", line 254, in _run_main
sys.exit(main(argv))
File "/content/drive/MyDrive/TheAIGuy/yolov4-custom-functions-master/detect.py", line 136, in main
image = utils.draw_bbox(original_image, pred_bbox, FLAGS.info, allowed_classes=allowed_classes, read_plate = FLAGS.plate)
File "/content/drive/MyDrive/TheAIGuy/yolov4-custom-functions-master/core/utils.py", line 259, in draw_bbox
cv2.rectangle(image, c1, (np.float32(c3[0]), np.float32(c3[1])), bbox_color, -1) #filled
cv2.error: OpenCV(4.7.0) :-1: error: (-5:Bad argument) in function 'rectangle'
> Overload resolution failed:
> - Can't parse 'pt2'. Sequence item with index 0 has a wrong type
> - Can't parse 'pt2'. Sequence item with index 0 has a wrong type
> - Can't parse 'rec'. Expected sequence length 4, got 2
> - Can't parse 'rec'. Expected sequence length 4, got 2
when running
python detect.py --weights ./checkpoints/custom-416 --size 416 --model yolov4 --images ./data/images/car2.jpg --plate
Plz help me with it
Thank you
I spend several minutes reading comments, to my surprise, you don't only pay much attention on creating good content but also you do well to answer comments. I hate those who just dump tutorials and don't even bother to attend to viewers.
Keep the good work.
Greetings from Tanzania 🇹🇿
Appreciate the kind words. Thanks and wishing you the best.
Dude....You're awesome
I think your thresholds to determine if it is a Number bounding box or not is very crucial. If you have a big image or a very small Number plate of a far away car, than this methods might fail. I know this is only a tutourial and no industry project :)
Thanks for the feedback. I appreciate it a lot. :)
Can we have a video for vehicular speed detection using YOLOv4 in TensorFlow...?
For me, it doesn't detect the actual number plate final value, it just highlights the number plate and gives a value, say 0.99 but the string where the number plate should have been there is missing.
Same happens during a video as well
same
Hello, thank you for the video. I wanted to ask you, is there a simple method to alternate the command for detection, e.g.: (python detect.py --weights ./checkpoints/custom-416 --size 416 --model yolov4 --images ./data/images/car2.jpg --plate) to use it on the whole folder? E.g. -images ./data/images/*.jpg (did not work)?
Did this command work for you?
thanx mate, I am a fan. but you have left us with a job on downloading the binaries. I have installed the program on windows, but the links takes you somewhere far to download. Id you could have filtered your video and take time to show us that, as the other part are explained well already. I am working on it. I does recognition but does not extract as per your video. Looking forward for more videos
It was a quick fix, to add it to the path environment, if anyone get stuck on a windows machine check this video th-cam.com/video/Rb93uLXiTwA/w-d-xo.html
Wow, it feels like you have just read my mind! I've been working on this project for some time now but the tesseract ocr part is a little bit complicated for me... I've tried pytesseract as you did, but the font used in license plates in my country confuses it and the error is high. So I've been trying to finetune the tesseract with my font and well.. it has been quite a challenge. Thank you for the great content as always!
Thanks for the kind words!
I agree with you on that..Tesseract does always work perfectly. .Another way would be to use a cloud API like AWS detect_text() function which performs better. But again, you want a local solution and nothing to do with cloud I presume.
I had madee this project before de lanch of yolov3, now will use the v4
@@rubyonrail1875 not working with cloud computing but that's a good suggestion. thank you!
@Isabella Fukushima, I just realised that what I meant to say in my previous message was that "Tesseract does NOT always work perfectly"..Just a point of correction.
😊
Amazing video. It's possible to use this for OCR in old pdf files, with low quality image?
It's possible to first detect the font?, like the web WhatTheFont.
After detecting the font, comparing the original image with the original font to detect the letters reducing the image proccessing and filters to the image?
I tried to use the software Rescribe, but the detection is not good enough for old scanned books.
Thank you for sharing all this. Greetings from Argentina.
Thanks @AIGUY for delivering great content.
Wow thanks, much appreciated. All the best :)
So I’ve been watching your vids and created a program that recognizes digits on a card. 4 digits in my case. I’m to “pass” the recognized digits to another program that will handle some math calculations and solve a math problem for me. Can I use this video to learn how to use/pass/ assign “recognized” digits to a variable of some sort for it to be later used? Hope my question makes sense!
Yes you could modify this program to read the digits using tesseract OCR for the image or you could learn from this code how to pass a detection on to a custom function! All the best!
The AI Guy Thanks man!!
@@TheAIGuy Do you think it would be better to just use your other video?. I just found it. How to Build Object Detection APIs Using TensorFlow and Flask
Hello, I've been following your tutorials, they're really awesome. I've been working on this project since 2 weeks and I've tried several settings of Tesseract but it isn't working as expected. Because of skew and other problems, I haven't tried your code yet. But anyway, keep making such good content.
Appreciate and hope this project helps your efforts!
@@TheAIGuy yes, I'll try this code, as soon as I can. And thanks once again for spending so much efforts to create such amazing content.
5:50 When I enter the command, I get an error --> ValueError: cannot reshape array of size 4554552 into shape (1024,512,3,3)
Can you teach me how to solve it?
Hello,
I found the video very interesting. I followed all the steps in order. However, when I execute the last command, I only get the image with the license plate circled in red, but without the recognition of the different characters. Would you know how to fix this problem please?
Why I can't run this on Google Colab ??
Thank you so much! I am so new to this. I am an old A.V. nerd. I have just started experimenting with Blue Iris and Code Project. I get okay results. I was wondering if I could implement this in Blue Iris for detecting license plates, since Code Project is so wonky?
I have an error when i use
"python save_model.py --weights ./data/custom.weights --output ./checkpoints/custom-416 --input_size 416 --model yolov4"
in the screen appear "DLL load failed while importing _pywrap_tensorflow_internal: The specified module could not be found"
how to fix this error, help me pls!
Thanks Guy for the videos. These are easy to follow and works perfect.
Just want to know if we can add new my own class to existing yolov4 classes. So that I don't need to retrain whole model from scratch?
Sir,Can we run this code on cuda 11.4 version or not? because i was facing an error on ubuntu "
ERROR: Could not find a version that satisfies the requirement tensorflow-gpu==2.3.0rc0 (from -r requirements-gpu.txt (line 1)) (from versions: 2.2.0, 2.2.1, 2.2.2, 2.2.3, 2.3.0, 2.3.1, 2.3.2, 2.3.3, 2.3.4, 2.4.0, 2.4.1, 2.4.2, 2.4.3, 2.4.4, 2.5.0, 2.5.1, 2.5.2, 2.5.3, 2.6.0, 2.6.1, 2.6.2, 2.6.3, 2.6.4, 2.6.5, 2.7.0rc0, 2.7.0rc1, 2.7.0, 2.7.1, 2.7.2, 2.7.3, 2.8.0rc0, 2.8.0rc1, 2.8.0, 2.8.1, 2.8.2, 2.9.0rc0, 2.9.0rc1, 2.9.0rc2, 2.9.0, 2.9.1)
ERROR: No matching distribution found for tensorflow-gpu==2.3.0rc0 (from -r requirements-gpu.txt (line 1))
" while installing " pip3 install -r requirements-gpu.txt " command.
Accuracy is very low . How to increase Accuracy ??
Hello!
I am trying to run the following command on my command prompt pip install -r requirements-gpu.txt but it can not find tensorflow-gpu 2.3.0cr , I tried to download the tensorflow-gpu but it says failed building wheel for tensorflow GPU, is that because my graphics card is intel 520 ?
Hi, I have follow your tutorial step by step, but whenever i run the command license plate recognition only the box show up but the result plate text didn't. Am I missing some step?but I'm sure doing exactly the same thing in the video, please help me
i'm dealing with the same problem, did u find the solution ?
@@michaelyip309 I did, turns out that I need to set PATH for tesseract on my machine and restart my pc, that worked for me
@@rizkijunianto846 will try that, thank you sir 👌
Hello, @AIGuy hope you are doing well. Can you make a video related to PP-YOLO and how PP-YOLO is different from YOLOv4?
Thanks !!
python save_model.py --weights ./data/custom.weights --output ./checkpoints/custom-416 --input_size 416 --model yolov4
bro when i run this command using i'm getting a pop up like "PYTHON HAS STOPPED WORKING"
can you please help me out to solve this issue
i am using CPU for tensorflow
bro please help me out
thanks for your job,can you explain this:
download the binary files and set them up on your local machine
The link of YOLOv4 weights is down! please re-upload it
Where can I download your dataset, images and corresponding labels?
Following your tutorial directly but gives errors after trying to load the custom .weights
I suppose after getting more subscribers then you can transfer to paid service.
Good work!!!!!!!!! but can you tell me how to run this model in cloud?
Vehicle detection using yolov5 (latest version)
When I converted my own custom model to tensorflow it doesnot show any detections
can you please help
The LPR is ambiguous at least for AT and DE plates. If you don't recognize the stickers you will fail. "S TA 5131E" will be the same as "STA 5131E", but those are different cities / regions. True for literally any German plate.
"Contours which have no meaning for us"... in fact, they have. They separate cities/regions from the rest of the plate
Anyone can reply me, why this doenst use epoch?
I don’t have a gpu
How to runs this in colab ?
anyone tried to implement this on flask? :D
After executing all your steps step by step in the Windows10 environment after I execute this command at the very end of the videos
"python detect.py --weights ./checkpoints/custom-416 --size 416 --model yolov4 --images ./data/images/car2.jpg --plate"
He returns this error to me here
"
Traceback (most recent call last):
File "detect.py", line 146, in
app.run (main)
File "C: \ ProgramData \ Anaconda3 \ envs \ yolov4-cpu \ lib \ site-packages \ absl \ app.py", line 300, in run
_run_main (main, args)
File "C: \ ProgramData \ Anaconda3 \ envs \ yolov4-cpu \ lib \ site-packages \ absl \ app.py", line 251, in _run_main
sys.exit (main (argv))
File "detect.py", line 49, in main
saved_model_loaded = tf.saved_model.load (FLAGS.weights, tags = [tag_constants.SERVING])
File "C: \ ProgramData \ Anaconda3 \ envs \ yolov4-cpu \ lib \ site-packages \ tensorflow \ python \ util \ deprecation.py", line 324, in new_func
return func (* args, ** kwargs)
TypeError: load () missing 1 required positional argument: 'export_dir' "
Could you help me urgently is the last part missing: /
Can we use the custom weights in yolo v7?
can we use this in colab brother
Thank you for this tutorial
Hello Sir! I follow all your tutorials. I did this project and integrated with PYQT5. I was doing number plate detection on cpu which is extremely slow. Now I bought new laptop with graphic card. I paste same project in new laptop.
It is detecting 1st frame only on video with gpu. I can't figure out what to do? Can you please help me with that, I want to display project next week!!
Excellent work!
following the exact steps, I received reshape error when running "python save_model.py --weights ./data/yolov4.weights --output ./checkpoints/yolov4-416 --input_size 416 --model yolov4"
utils.load_weights(model, FLAGS.weights, FLAGS.model, FLAGS.tiny)
File "C:\Users\mike\car_tracking\src\yolov4-custom-functions\core\utils.py", line 143, in load_weights
conv_weights = conv_weights.reshape(conv_shape).transpose([2, 3, 1, 0])
ValueError: cannot reshape array of size 4554552 into shape (1024,512,3,3)
I have a serious problem and question please how can i reach you?i can detect licence plates now but now i want to process the image and extract the text i have used your darknet version but the tutorial that i see you have extracted text text is in tensorflow pls how do i process the image detected in darkent?Please i am so stuck here help me
Dude !
Can the tesseract use a video in place of picture to read the number plate ?
As far as I'm aware Tesseract can only run on images, so that is why you need to take a frame of the video as the image input to tesseract. If there is an easier way then I'd love to know!
Can we run this on google collab?
i have a error when Running License Plate Recognition on Images
tensorflow.python.framework.errors_impl.InvalidArgumentError: ConcatOp : Dimensions of inputs should match: shape[0] = [1,0,1] vs. shape[4] = [1,1,0]
[[{{node StatefulPartitionedCall/functional_1/tf_op_layer_concat_18/concat_18}}]] [Op:__inference_signature_wrapper_5589]
help me
Hi mate ,thanks for the amazing video. How can I save the predicted coordinates in a "Result.txt" file, like we are doing earlier ?
while running # custom yolov4 function its showing error . ValueError: cannot reshape array of size 72859 into shape (512,256,3,3) can any one help
You are a god
haha thanks so much.
Hi, I have question. my plate detector is not working. it can only detect the plate with a bounding box. I have installed the tesseract-ocr for windows. I have also installed all the library in the requirements.txt. Do you have an idea what is the possible problem? thanks.
Hi! One question, I have a custom trained model of YOLOv4 but I run some prunning for better perfomance but now I can't convert to tensorflow. Any ideas? Thank you!
Hello I'm Oscar
And I'm Working on car plate recognition can I get help from you dear it's my final year project
Great job and explanation. I learnt a lot with this tutorial and others (deepsort one). Deepsort works nice on cars with yolov4-tiny. I'm wondering if it would be possible to train this custom plate detector for yolov4-tiny and make it faster?
Could it run in realtime?
Really great stuff!!! I have one question... when you have the tensorflow model... can you the also train the model in tensorflow?
Yes there are ways to train the model within TensorFlow itself. I always find training YOLOv4 in Colab easiest however.
Running well.But NOT GETTING ANYTHING FINALLY.............anyone help
What to do with video..
hey,not sure if you're gonna reply to this but, do you have any advice on if the license plates im reading are on black backgrounds and on white texts instead?
dude you are awesome literally working on this project it helps a lot. and my question is how do i train on all vehicle like bikes,trucks etc
If you are looking to train a custom YOLOv4 model on those classses then check out one of my previous videos that goes over how to do exactly that!
@@TheAIGuy hi! i trains the custom yolov4 model on indonesian number plates and its showing great results could you tell me how --crop parameter works because the video detect many plates but the code does't crop many of them could you please tell me how i set its threshold value
@@talhayaqoob889. hi. The same problem here. Have you found any solution?
@@senolkurt7864 the problem is not with the vehicles there are many types you should have train model on custom plates because plates may be vary in size like on bikes and cars .
@@talhayaqoob889 . my problem is it detects the plates in a video but doesn't crop and save those images in "detections" folder.
Hi it has worked but i want to run it on webcam which command do i run for that?or video
Hey what if I want to do text extraction on another image like ID_CARD with 2 or more classes , can I use the same method and can I use yolov3 instead of yolov4.
As I already trained my data set.
thanks so much man, im doing my final year of university and im doing something similar for my final project, im still a bit confused by this because i never really used python but it was very helpful
'aws' is not recognized as an internal or external command,
operable program or batch file.
For the people who are getting this error try installing aws cli on your local system and re-run your command.
if the error persists try using command prompt instead of PowerShell.
someone help me which command do i run to detect from a webcam please?
Plz reply, can you please tell a way to package all of this in a exe sort of thing which can be uses as a command line utility for easy distribution rather than installing python and all requirements on every system I wanna use it on.
Exelent work btw, Thanks a lot
Hi, I have a question, is there any way to save the plate result into csv file, should I make the save function separately or can I put into some function that already in the project?please I really need this done by the end of the month
Can you make a video about training a custom data ( Yolov3 or Yolov4) using google colab? Thanks
謝謝
Bro, can u make a video of implementation in colab?
if u did it, kindly share the link.
@@RahulKumar-oy3pd Follow the procedure in your physical laptop because the trained model can be downloaded and hence does should not require much RAM.
Yeah,that's fine.but wanted to do it on collab for further deployment of the model.
Hey great content and delivery man! Just had a question. I did everything as instructed and can run the commands to the point where the license plate is detected in pictures. The only mishap is that I am not getting License Plate # output on my console. Do you have an idea on what the problem may be?
The tesseract is not installed properly, I had the same problem. Make sure after installing it, you set it to the system variable path.
@@yaseenyaseen24 Thank you!
@@gabrielacuna8530 did you get the solution?
Just subscribed before following the video this is what have been looking for!
Hi thanks @The AI Guy, When I detected license with more than 1 line, It can't be recognized. Please provide how can I improve the code to recognize 2 lines? Thank you.
it tells me no module named cv2 when trying this line.. python save_model.py --weights ./data/yolov4-tiny.weights --output ./checkpoints/yolov4-tiny-416 --input_size 416 --model yolov4 --tiny
cant solve it
CUDA runtime implicit initialization on GPU:0 failed. Status: device kernel image is invalid also this. maybe I have a problem with my cuda?
Helloo great work ! I wanted to ask if I can train YOLOV4 on my own dataset to detect individual digits in the images and then recognize the digits with the customized function you used ?
bro how to clone repository in windows, when i type "git clone xyz link " its showing git is not internal or external command
You probably don’t have git installed on your machine. You can install git and then try again or you can hit the “Download ZIP” button to download the code.
@@TheAIGuy haa I download it now thank you for the response, now I got one new error, for while I'm following ur video point to point, I don't have tensor flow how to download it??
When I type conda command in git shell it's showing conda command not found
@@Aadil-Khan22 Try typing the conda command in the anaconda shell
👍👍👍 Good Job , Gentleman
Appreciate it!
Guys custom names ka folder anyone can guide where it is ?
Which camera did you use for the number plate recognition
Hi . To recognize license plates from images located at a large height. Does this same tutorial work?
hi man , did this tuturials works withs an AMD GPU ?
What if we trained another YOLO on dataset for digital/printed digits, no need for anything classical.
i have applied the same preprocessing steps but i get a only white images
Amazing explanation, amazing repo. Sad the recognition is pretty slow, for real usage might be to slow, as for detecting plates in a video feed.
Thank you very much for this great content. I have a question : is this proposed method work at night and confirm it the robustness against severe imaging conditions?
Could you, please, make a tutorial for Real-time license plate recognition based on improved Tiny-YOLOv3 ? n morning and night if that possible
Last question there is any tutorial about segmentation of characters in Arabic license plate?
Thank you in advance and the best for you
Thank you very much, a question
How can I use it in real time with a webcam?
when I save the model into tensorflow format, the prediction becomes bad. I did this on google colab. And the tensorflow model's size is also way much smaller than the darknet one
Thank you for this informative tutorial @The AI Guy. I want to ask about dataset. Can you share it with me? I'm also going to train YOLOv4 on custom dataset. It will help me a lot. Or you can tell me how you labeled your dataset and how much classes you have? I'm waiting of your reply. Thank you!
Very helpful video . Thanks 😃
You deserve 10M Subscriber
Awesome tutorial as always!! Amazing!!
I didn't know that Danny Amendola knew so much about AI.
Hey, I liked your video just want to know can we do this by uploading a bulk of images in testing like 3 to 4 car images together?
anyway, I can tweak some parameters to make it more accurate, currently, it's only giving wrong predictions?
can save output plate number in database how?
First of all, thx for ur work.
Because of you i started with python. And now i try to learn more about DL and object detection.
I wanna test out how much slower the object detection on my GPU is in comparision to the GPU from Google Colab.
In your Colab tutorials u implemented a time measurment for the object detection.
Can u give me a advice how i can implement such a thin in this --info function?
Hi! Im verry impressed by your coding and your overall knowlege and I have to say that am I a complete noob is this area. Im wonderying if you can compile all of that code and turn it into a program that you can simply run by clicking on a executable? Let me know, looking foward to ear from you!