Automatic Number Plate Recognition

Group homework for the 7th semester class Image Processing at the Budapest University of Technology and Economics.

• project-summary.pdf

Dependencies

Install Python packages with pip:

pip install -r requirements.txt

TLDR Usage

• Put images in the images folder.
• ./detect.sh
• output_original.csv contains the results (in the root folder)
• output.csv contains the results of each cutout
  • good for debugging purposes, and matching these against the detections in images/cutouts/

Usage

• Put images in the images folder.
• Run YOLO object detection
  • python yolov7-number-plates/detect.py --weights yolov7-number-plates/yolov7-number-plates-trained.pt --img-size 448 --source images --name number-plates-yolo --save-txt --save-conf --nosave --project images --exist-ok
• Cut out detections
  • move detected *.txt labels into folder images/labels
    • mv images/number-plates-yolo/labels images/labels
  • python cutout.py images
  • Results are now in the images/cutouts folder
• Run OCR on detections
  • python paddleOCR/main.py
• output.csv contains the results (in the root folder)
• Run python fix_csv.py to merge detections from the same image to the same line, with the original file names
• The final results are in output_original.csv

Individual Components

• License Plate Object Detection: https://github.com/bobarna/yolov7-number-plates
• OCR: we implemented 2 different methods, and ended up using paddleOCR
  • paddleOCR: paddleOCR folder
  • custom OCR: LicensePlateOCR folder

Building the Documents Locally with LaTeX

See docs folder.