About integrating OCR with Yolo Model

Discussion
,
1

So currently i am a junior ml engineer. Right now I am doing a tricky project on ship container number detection using OCR. I have seen paddle ocr doing great results. But for the crop in challenging lighting condition and overall i need to annotate each text within the number region. Now give me your thoughts how should i proceed. If anyone has done such tricky projects, let me know. Thank you so much. Also some of the containers are less likely visible where i need to use check_digit. Give me overall idea about it. Also i will integrate this inside in a jetson edge device. The idea is whenever the slacker is going to pull up a container it will detect the text region and it will recognize the code inside it and it will also send the overall gps with the code and time.

1 Like
2

Hi there! :blush:

Integrating OCR with a YOLO model for ship container number detection sounds like an exciting project! Here’s a step-by-step approach you might find helpful:

  1. Model Selection: Since you’re dealing with challenging lighting conditions, YOLO models are a great choice for detecting text regions due to their real-time performance. You can use YOLO to detect the container number regions and then pass these regions to an OCR model like PaddleOCR for text recognition.

  2. Annotation: You’ll need to annotate your dataset with bounding boxes around the text regions. Tools like LabelImg can be useful for this task. Ensure your annotations are precise to improve model accuracy.

  3. Check Digit Verification: For containers that are less visible, implementing a check digit algorithm can help verify the accuracy of detected numbers. This involves calculating a checksum based on the detected digits and comparing it to the check digit.

  4. Integration on Jetson: Deploying on a Jetson device is a great choice for edge applications. Make sure to optimize your models for inference on Jetson, possibly using TensorRT for acceleration.

  5. Data Collection and Annotation: For more detailed guidance on data collection and annotation, check out our Data Collection and Annotation Guide.

  6. GPS and Time Integration: You can use a GPS module to capture location data and combine it with the detected text and timestamp for a comprehensive output.

Feel free to explore our Ultralytics documentation for more insights. Best of luck with your project, and feel free to share your progress! :rocket:

If you encounter any issues, make sure to check if they persist with the latest package versions. Happy coding!

1 Like