🚦 SpeedRadar-ANPR

AI-Powered Speed Enforcement & Automatic Number Plate Recognition System

An end-to-end Computer Vision pipeline for automated traffic monitoring, speed enforcement, and automatic license plate recognition.

The system detects and tracks vehicles in real-time, estimates their speed, identifies speed violations, extracts license plate information using OCR, generates visual evidence for offenders, and automatically produces violation reports.

---

📸 Screenshot

The screenshot above shows a completed run of the pipeline on the sample traffic video — the live terminal output detecting each violation in real time (speed, plate, fine) alongside the final auto-generated `report.md`, which lists all 5 detected violations with their speed, plate number, fine amount, and a link to the corresponding proof image.

---

📑 Table of Contents

• Features
• Architecture
• Demo
• Installation
• Usage
• Project Structure
• Challenges & Solutions
• Future Improvements
• Technologies Used
• License
• Disclaimer
• Contact

---

🌟 Features

🚗 Real-Time Vehicle Detection & Tracking

• Detects vehicles using YOLO11.
• Tracks vehicles across frames using ByteTrack.
• Assigns a unique ID to every detected vehicle.

📏 Dynamic Velocity Estimation

• Calculates vehicle speed using pixel displacement and frame timing.
• Converts image-space movement into real-world speed estimates.
• Uses motion filtering to eliminate speed jitter caused by stationary vehicles.

🎯 Proximity-Aware OCR

• Runs the license plate recognition pipeline only when the vehicle is sufficiently close to the camera.
• Reduces OCR failures caused by distant or low-resolution plates.
• Improves overall ANPR accuracy.

🔁 Multi-Frame OCR Retry Mechanism

• Attempts OCR across multiple frames if the initial read fails.
• Uses Regex filtering to validate plate formats.
• Selects the highest-confidence plate reading.

💰 Automated Fine Calculation

• Dynamically calculates fines according to configurable speed limits.
• Supports multiple violation brackets.

📸 Watermarked Visual Evidence

Automatically generates evidence images containing:

• Vehicle snapshot
• Measured speed
• License plate number
• Fine amount
• Timestamp

🏷 Plate-Named Evidence Storage

Evidence images are automatically saved using the detected license plate number as the filename for fast lookup and traceability.

Example:

proof/
├── ABC1234.jpg
├── XYZ5678.jpg
└── DEF9012.jpg

📄 Automatic Violation Reports

Generates a Markdown report containing:

• Vehicle ID
• License plate number
• Vehicle speed
• Speed limit
• Fine amount
• Total revenue generated from fines

---

🏗 Architecture

Traffic Video
      │
      ▼
YOLO11 Vehicle Detection
      │
      ▼
ByteTrack Tracking
      │
      ▼
Speed Estimation
      │
      ▼
Speed Limit Check
      │
 ┌────┴────┐
 │Violation│
 └────┬────┘
      ▼
License Plate Detection
      ▼
PaddleOCR Recognition
      ▼
Regex Validation
      ▼
Fine Calculation
      ▼
Evidence Generation
      ▼
Markdown Report

---

🎥 Demo

The repository includes a sample traffic video for quick testing:

traffic_video/
└── sample_video.mp4

Simply run:

python main.py

and the system will process the included sample video automatically.

The repository also contains:

assets/
└── photo.png

which showcases an example of the generated visual evidence produced by the system.

A complete demonstration video showing:

• Vehicle Detection
• Vehicle Tracking
• Speed Estimation
• License Plate Recognition
• Fine Calculation
• Evidence Generation
• Report Creation

is available here:

https://drive.google.com/drive/folders/1RB6IiRvOlSUCQAtKY5DU89HfoJKRaABN?usp=sharing

Example outputs:

output/
└── output_video.mp4

proof/
├── ABC1234.jpg
├── XYZ5678.jpg
└── DEF9012.jpg

report.md

---

⚙️ Installation

Prerequisites

• Python 3.10+
• CUDA GPU (Optional but recommended)

Clone Repository

git clone https://github.com/usfahmed/SpeedRadar-ANPR.git
cd SpeedRadar-ANPR

Install Dependencies

pip install -r requirements.txt

Install PaddlePaddle

CPU Version

pip install paddlepaddle

GPU Version

Install the version compatible with your CUDA installation.

---

🚀 Usage

The repository already includes a sample traffic video:

traffic_video/
└── sample_video.mp4

You may replace it with your own footage if desired.

Run the application:

python main.py

Enter the desired speed limit when prompted:

Enter speed limit: 60

After processing is complete:

📁 output/

Contains the fully annotated processed video.

📁 proof/

Contains watermarked evidence images named using the detected license plate number.

Example:

proof/
├── ABC1234.jpg
├── XYZ5678.jpg
└── DEF9012.jpg

📄 report.md

Contains the generated violation report.

---

📂 Project Structure

SpeedRadar-ANPR/
│
├── assets/
│   └── photo.png
│
├── detections/
│   ├── __init__.py
│   ├── car_detection.py
│   └── licence_plate_detection.py
│
├── models/
│   ├── best.pt
│   └── yolo11n.pt
│
├── output/
│
├── proof/
│
├── traffic_video/
│   └── sample_video.mp4
│
├── utils/
│   ├── __init__.py
│   └── video_utils.py
│
├── LICENSE
├── main.py
├── README.md
└── requirements.txt

---

⚠ Challenges & Solutions

Problem

Stationary vehicles occasionally generated false speed readings because of tracker jitter.

Solution

Implemented motion-distance filtering before speed calculation.

---

Problem

OCR accuracy dropped significantly for distant vehicles.

Solution

Implemented proximity-aware OCR triggering based on bounding box size.

---

Problem

Motion blur occasionally caused OCR failures.

Solution

Implemented multi-frame OCR retries with confidence scoring and Regex validation.

---

🔮 Future Improvements

• Multi-camera support
• Real-time dashboard
• Database integration
• REST API support
• Cloud deployment
• Vehicle classification
• Traffic statistics dashboard
• Heatmap generation
• Email notifications
• SMS notifications

---

🛠 Technologies Used

Programming Language

• Python

Object Detection

• YOLO11

Multi-Object Tracking

• ByteTrack

OCR

• PaddleOCR

Computer Vision

• OpenCV
• NumPy

Data Processing

• Pandas

Reporting

• Markdown

---

📄 License

This project is licensed under the MIT License.

See the LICENSE file for more details.

---

⚠ Disclaimer

This project was developed for educational and research purposes only.

It is not intended for deployment in real-world law enforcement systems without proper calibration, validation, legal approval, and regulatory compliance.

---

📫 Contact

Youssef Ahmed Abdelfatah

🌐 Portfolio
https://usfahmed.dev

💻 GitHub
https://github.com/usfa7med

💼 LinkedIn
https://linkedin.com/in/usfa7med

✉️ Email
hello@usfahmed.dev