🚦 Intelligent Traffic Surveillance System (ITSS)

Automating Traffic Compliance in India using Attention-Centric Real-Time Object Detection.

---

📖 Table of Contents

• Abstract
• Key Features
• System Architecture
• Methodology & Logic
• Tech Stack
• Performance Evaluation
• Installation
• Usage
• References & Citations

---

📄 Abstract

India’s rapid urbanization has led to a surge in vehicular density, rendering traditional manual traffic monitoring inefficient. Manual oversight is prone to human error, corruption, and inability to handle high-volume traffic 24/7.

ITSS (Intelligent Traffic Surveillance System) is a vision-based automated solution designed to enforce traffic laws with high precision. By leveraging the YOLOv12 (You Only Look Once) architecture, which introduces attention-centric mechanisms for superior real-time performance, this system detects complex violations such as red-light jumping, speeding, and helmet-less riding.

Furthermore, the system integrates an OCR (Optical Character Recognition) pipeline to extract license plates and communicates with a backend API to verify Insurance and Pollution Control (PUC) status, automatically generating e-challans for offenders.

---

✨ Key Features

1. Visual Violation Detection

• 🔴 Red Light Violation: Detects the traffic signal state and identifies vehicles that cross the stop line while the signal is red.
• ⛑️ Helmet-less Rider Detection: Classifies two-wheeler riders and specifically detects the absence of a helmet on both the rider and pillion.
• ⚡ Speeding Detection: Uses perspective transformation and object tracking to estimate vehicle speed in real-time.

2. Automatic Number Plate Recognition (ANPR)

• 🔍 Plate Extraction: High-accuracy cropping of license plates from moving vehicles.
• 📝 OCR Processing: Converts plate images to text strings using advanced OCR engines.

3. Compliance & Enforcement

• ☁️ API Integration: Queries a central database using the extracted plate number.
• 🚫 Document Verification: Checks for valid Insurance and PUC certificates.
• 📩 E-Challan Generation: Automatically creates a violation report with time, location, violation type, and evidence image.

---

🏗 System Architecture

The system operates on a pipeline approach:

1. Input Acquisition: CCTV feed or pre-recorded video.
2. Object Detection (YOLOv12): Detects classes: Vehicle, License Plate, Traffic Light, Person, Helmet.
3. Object Tracking: Assigns unique IDs to vehicles across frames using SORT/DeepSORT to handle occlusion.
4. Violation Logic Module:
   • If Red Light AND Vehicle Center > Stop Line → Violation.
   • If Vehicle Speed > Speed Limit → Violation.
   • If Motorbike AND Head AND No Helmet → Violation.
5. Post-Processing:
   • Crop License Plate → Pass to OCR → Get String.
   • Send String to API → specific JSON response.
6. Output: Overlay visuals on video and log data to CSV/Database.

---

⚙️ Methodology & Logic

Why YOLOv12?

We utilize YOLOv12 by Ultralytics. Unlike previous iterations, YOLOv12 introduces an attention-centric architecture that significantly improves feature extraction in complex urban environments (e.g., crowded Indian roads). It balances the speed of CNNs with the global context awareness of Transformers.

• Reference: Ultralytics YOLOv12 Documentation

Speed Estimation Logic

To calculate speed from a 2D video feed:

1. Perspective Transform: We map the Region of Interest (ROI) on the road to a "bird's eye view."
2. Euclidean Distance: Calculate the distance moved by the vehicle centroid between frames.
3. Formula: $Speed = \frac{Distance (meters)}{Time (seconds)} \times 3.6 (km/h)$

---

🛠 Tech Stack

Component	Tool/Technology	Description
Language	Python 3.9+	Core logic and scripting.
Detection Model	YOLOv12	Custom trained on Indian Traffic Dataset.
Framework	PyTorch / Ultralytics	Model training and inference.
Computer Vision	OpenCV	Image processing, perspective transforms.
OCR	EasyOCR / Tesseract	Text extraction from license plates.
Tracking	DeepSORT	Multi-object tracking for vehicle persistence.
Backend	Flask / FastAPI	Handling API requests for challan generation.

---

📊 Performance Evaluation

The model was tested on a custom dataset comprising diverse Indian road scenarios (day, night, rain, high density).

• Overall Accuracy: 93.76%
• mAP@0.5: 0.95
• Inference Speed: ~45 FPS on NVIDIA RTX 3060
• OCR Accuracy: 89% (Dependent on plate visibility)

---

📦 Installation

Prerequisites

• Python 3.8 or higher
• CUDA-enabled GPU (Recommended for real-time performance)

Steps

1. Clone the Repository

   git clone [https://github.com/yourusername/traffic-surveillance-system.git](https://github.com/yourusername/traffic-surveillance-system.git)
   cd traffic-surveillance-system

2. Create Virtual Environment (Optional but recommended)

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

   # Install Ultralytics for YOLOv12
   pip install ultralytics
   
   # Install other requirements
   pip install -r requirements.txt

4. Download Weights Place your trained yolov12_custom.pt file in the weights/ directory.

---

💻 Usage

To run the system on a video file:

python main.py --source data/input_video.mp4 --weights weights/yolov12_custom.pt --conf 0.5

Arguments:

• --source: Path to video file or 0 for webcam.
• --weights: Path to the trained YOLOv12 model.
• --conf: Confidence threshold for detection.
• --save-txt: Save violation logs to a text file.

---

🔗 References & Citations

This project is built upon the cutting-edge research in object detection. If you use this repository or the YOLOv12 architecture in your research, please cite the original authors:

YOLOv12 Architecture

@article{tian2025yolo12,
  title={YOLOv12: Attention-Centric Real-Time Object Detectors},
  author={Tian, Yunjie and Ye, Qixiang and Doermann, David},
  journal={arXiv preprint arXiv:2502.12524},
  year={2025}
}

YOLOv12 Software

@software{yolo12,
  author = {Tian, Yunjie and Ye, Qixiang and Doermann, David},
  title = {YOLO12: Attention-Centric Real-Time Object Detectors},
  year = {2025},
  url = {[https://github.com/sunsmarterjie/yolov12](https://github.com/sunsmarterjie/yolov12)},
  license = {AGPL-3.0}
}

Acknowledgements

• Ultralytics: For the YOLO framework implementation. Docs
• OpenCV: For image processing tools.

---

_{Developed with ❤️ for safer roads in India.}