Vehicle Object Detection and License Plate Recognition (LPR) System

What is AegisLens?

AegisLens is a model that can detect vehicle types and read their license plate. It utilizes an object detection model and an Optical Character Recognition (OCR) model to output it’s results.

How does it work?

1. Object Detection: The model detects vehicles in the video footage and identifies their bounding boxes.
2. License Plate Recognition: For each detected vehicle, the model extracts the license plate region and applies OCR to read the text.
3. Expiry Date Extraction: The model processes the license plate text to extract the expiry date, if available.
4. Output: The model displays the detected vehicle type, license plate number, and expiry date on the video footage.

Tools for model development:

• YOLOv8: A object detection model used for detecting vehicles in the video footage.
• EasyOCR: An OCR model used for reading text from the detected license plates.
• Streamlit: A web application framework used to create the user interface for the application.

Features:

• Exploratory Data Analysis: View insights and statistics about the dataset used for training the model.
• Prediction: Upload a video file to detect vehicles and read license plates.

Repository Overview

.
├── AEGISLENS_EDA_ENG.ipynb           # Exploratory Data Analysis in English
├── AEGISLENS_EDA_ID.ipynb            # Exploratory Data Analysis in Indonesian
├── Archive/                          # Archived files or previous versions
├── data.yaml                         # YOLO dataset configuration file
├── demo1.avi                         # Sample demo video for testing
├── Deployment/                       # Scripts or configs related to model deployment
├── inference.py                      # Inference script using trained YOLO model
├── latest.pt                         # Trained YOLO model weights
├── metrics/                          # Evaluation metrics and visualizations
├── ocr_plate.ipynb                   # OCR pipeline for plate recognition
├── README.md                         # Project overview and instructions
├── References/                       # External references or citations
├── requirements.txt                  # Python dependencies
├── splitting.py                      # Script to split dataset/video frames
├── take_frames_from_video.ipynb      # Notebook to extract frames from video
├── test.MOV                          # Test video sample
├── test2.MOV                         # Another test video sample
└── yolo_implementation.ipynb         # YOLOv5 implementation and training notebook

Project Description

This system combines YOLO (You Only Look Once) object detection with license plate recognition to:

1. Detect vehicles in images/video streams
2. Identify vehicle classes (car, truck, motorcycle, etc.)
3. Optionally recognize license plate details including expiration date

How to Run Inference

1. Initial Setup

Create and activate virtual environment:

python -m venv venv
source venv/bin/activate  # Linux/Mac
.\venv\Scripts\activate   # Windows

Install dependencies:

pip install -r requirements.txt

2. Running Detection

Basic Command Structure:

python inference.py \
    --model <model_path> \
    --source <input_source> \
    [--thresh <confidence_threshold>] \
    [--resolution <WxH>] \
    [--record]

3. Detailed Arguments

Argument	Description	Required	Default
--model	Path to YOLO model file (e.g., "runs/detect/train/weights/best.pt")	Yes	-
--source	Input source: - Image file ("test.jpg") - Image folder ("test_dir") - Video file ("testvid.mp4") - USB camera ("usb0") - Picamera ("picamera")	Yes	-
--thresh	Minimum confidence threshold (0.0 to 1.0)	No	0.5
--resolution	Display resolution in WxH format (e.g., "640x480")	No	Source resolution
--record	Record and save output as "demo.avi" (requires --resolution)	No	False

4. Usage Examples

Basic image detection:

python inference.py --model latest.pt --source test.jpg

Webcam detection with custom threshold:

python inference.py --model latest.pt --source usb0 --thresh 0.7

Video processing with recording:

python inference.py --model latest.pt --source traffic.mp4 --resolution 1280x720 --record

Batch process image folder:

python inference.py --model latest.pt --source images/ --thresh 0.6

5. Important Notes

1. For recording (--record), you must specify --resolution
2. Camera indices start from 0 (e.g., "usb0" for first USB camera)
3. Confidence threshold (--thresh) filters out low-confidence detections
4. Supported image formats: JPG, PNG, BMP
5. Supported video formats: MP4, AVI, MOV

6. Troubleshooting

Webcam not working:

• Try different indices ("usb0", "usb1", etc.)
• Verify camera permissions
• Check if OpenCV can access the camera:

  import cv2
  cap = cv2.VideoCapture(0)
  print(cap.isOpened())

Graphical User Interface (GUI) Application

Run the interactive desktop application with:

python UIApp/main_app.py

Installation

Prerequisites

• Python 3.8+
• NVIDIA GPU (recommended)
• CUDA/cuDNN (for GPU acceleration)

Setup

1. Clone repository:

   git clone https://github.com/yourusername/vehicle-plate-detection.git
   cd vehicle-plate-detection

2. Install dependencies:

   pip install -r requirements.txt

Applications

• Automated vehicle verification systems
• Parking management
• Traffic monitoring
• Insurance fraud detection
• Expired plate detection

Project Link: click here

Deployment Link: click here

License

This project is licensed under the MIT License - see the LICENSE file for details.

You must:

• Include original license
• Maintain copyright notice

No warranty is provided. Use at your own risk.

Contributors

• Lis Wahyuni - Data Scientist ([email protected])
• Faishal Kemal - Data Scientist & Data Annotator ([email protected])
• Muhammad Rafi Abhinaya - Data Scientist & Data Annotator ([email protected])
• Ma'ruf Habibie Siregar - Data Analyst & Data Annotator ([email protected])
• Fauzan Rahmat Farghani - Data Engineer & Data Annotator ([email protected])