parkocr 0.1.1

Automatic license plate recognition using YOLOv8 and EasyOCR, with ROI filtering, FIFO output, and screenshot support.

ParkOCR

ParkOCR is a Python library for automatic license plate recognition using YOLOv8 for detection and EasyOCR for text recognition.
It is designed to work with RTSP streams (e.g., IP cameras), webcams, or video files.

✨ Features

• YOLOv8 for plate localization (supports n, s, m, l, x models).
• EasyOCR for text extraction.
• ROI filtering: only detect plates inside a defined region.
• Headless mode (no window display).
• Prevents freezing on the same consecutive plate.
• Optional FIFO output for inter-process communication.
• Screenshot saving:
  • screenshot=None: disabled
  • screenshot="full": saves full frame
  • screenshot="roi": saves only the region of interest

📦 Installation

pip install parkocr

🚀 Usage

Basic example

from parkocr.detector import Detector

detector = Detector(
    rtsp_url="rtsp://user:password@192.168.0.100:554/onvif1",
    model_size=0,  # 0=nano, 1=small, 2=medium, 3=large, 4=xlarge
    headless=True,
    screenshot="roi",  # options: None, "full", "roi"
)

detector.run()

With FIFO output

detector = Detector(
    rtsp_url="rtsp://user:password@192.168.0.100:554/onvif1",
    fifo_output="/tmp/plates_fifo",
    headless=True
)
detector.run()

Now another process can read detected plates from /tmp/plates_fifo.

With callback

def on_plate(plate):
    print("Detected plate:", plate)

detector = Detector(
    rtsp_url="rtsp://user:password@192.168.0.100:554/onvif1",
    on_detect=on_plate,
    headless=True
)
detector.run()

⚙️ Parameters

• rtsp_url (str): RTSP stream, file path, or webcam index.
• model_size (int): 0=n, 1=s, 2=m, 3=l, 4=x. Larger = more accurate, slower.
• conf_thresh (float): YOLO detection confidence threshold.
• roi (tuple[int, int, int, int] | None): Region of interest (x1, y1, x2, y2). Defaults to central ROI.
• headless (bool): Disable display windows and overlays.
• window_size (tuple[int, int]): Window size for display.
• process_interval_s (float): Seconds between detection cycles.
• freeze_seconds (float): How long to freeze the display after detection.
• ocr_langs (list[str]): OCR languages for EasyOCR.
• on_detect (callable): Callback called with detected plate text.
• min_ocr_conf (float): Minimum OCR confidence.
• fifo_output (str | None): Path to FIFO file to send plates.
• screenshot (str | None): None, "full", or "roi".

📂 Project structure

parkocr/
├── detector.py
├── __init__.py
README.md
pyproject.toml

📝 License

MIT © 2025 João Luiz Thomazetti