Abstract

The Intelligent Speed Monitoring and Alert System for Traffic Management leverages the power of artificial intelligence to monitor and manage traffic violations efficiently. This project focuses on automating the detection of vehicles exceeding speed limits, capturing images of such violations, recognizing vehicle number plates, and notifying authorities via email. By utilizing advanced technologies such as PyTorch, YOLOv11, and OpenCV, along with the computational capabilities of NVIDIA DGX servers, this system achieves real-time performance and high accuracy. The solution addresses the challenges of manual traffic monitoring, making roads safer and improving compliance with traffic rules.

Introduction

Traffic management is a critical issue in modern urban environments, with over-speeding being one of the leading causes of road accidents. Traditional traffic monitoring systems often involve manual effort, which is time-consuming, error-prone, and inefficient.

The Intelligent Speed Monitoring and Alert System provides a fully automated solution by employing computer vision and machine learning techniques. The system is designed to:

** Calculate vehicle speed in real-time using video processing.
** Identify violations based on pre-defined speed thresholds.
** Capture frames of violating vehicles and recognize their license plates.
** Notify the concerned authorities via email, attaching evidence for further action.

This solution integrates seamlessly with existing infrastructures, offering a scalable approach to building smarter and safer cities.

Methodology

The development of this project involves multiple interconnected modules. Each module was carefully designed to ensure accuracy, efficiency, and reliability.

1. Vehicle Speed Detection

Speed is calculated by tracking the position of vehicles in consecutive frames of a video feed.

** Frame Analysis: OpenCV is used to process the video feed, track moving objects, and compute their displacement over time.

** Calibration: The system was calibrated to ensure accurate speed measurements by factoring in the video’s frame rate and physical distances captured.

2. Number Plate Recognition

After detecting a speed violation, the system extracts the vehicle’s number plate for identification. This process involves:

** Detection: Using the YOLOv11 model to locate number plates in the captured frames.

** Recognition: Optical Character Recognition (OCR) processes the detected plates to extract alphanumeric details.

3. Annotation Process

Annotation played a crucial role in training the YOLOv11 model. A custom dataset of vehicle images and videos was annotated meticulously to improve detection accuracy.

** Bounding Box Creation: Each image was manually labeled to define the exact regions containing vehicles and number plates.

** Class Categorization: Images were classified into specific categories such as "Car," "Bike," and "Truck" to train the model for varied scenarios.

** Dataset Size: Over 100 images were annotated, ensuring diversity in vehicle types, angles, and lighting conditions.

** Preprocessing: The annotated data was augmented (e.g., rotation, scaling) to simulate real-world conditions like night visibility and occlusions.

4. Email Alert System

The Gmail API was integrated to automate the notification process.

Each email includes:

• A high-resolution image of the violating vehicle.
• The extracted license plate details.
• A timestamp of the violation.

5. Deployment

The application was deployed on NVIDIA DGX servers to utilize GPU acceleration, ensuring real-time inference. Flask was employed as the backend framework for handling API requests and orchestrating different modules.

Experiments

The system was subjected to a rigorous testing process to evaluate its performance under various conditions:

1. Dataset Training

The YOLOv11 model was trained on the annotated dataset using PyTorch.

• Epochs: The training ran for 50 epochs with a batch size of 32.
• Loss Functions: Custom loss functions were implemented to minimize errors in detection.

2. Simulated Environments

The system was tested on video feeds simulating different traffic environments:

• Urban Roads: Mixed traffic with cars, bikes, and buses.
• Highways: High-speed traffic with vehicles moving at varying speeds.
• Low-Light Conditions: Night time video feeds to assess robustness under poor visibility.

3. Model Calibration

The vehicle speed detection system was calibrated by comparing computed speeds against ground truth measurements obtained manually.

Results

The system delivered high performance in both controlled and real-world scenarios:

1. Accuracy Metrics

• Vehicle Detection Accuracy: 97.5%
• Number Plate Recognition Accuracy: 95.3%

2. Real-Time Processing

The system achieved a processing speed of 29 FPS on NVIDIA DGX servers.

The email notification system had a 100% success rate, ensuring prompt delivery of violation reports.

3. Robustness

The system performed consistently well under various conditions, including low-light environments and heavy traffic scenarios.

Annotated data augmentation ensured that the model was robust to angle and size variations.

Conclusion

The Intelligent Speed Monitoring and Alert System for Traffic Management demonstrates the potential of AI to revolutionize traffic management systems.
By automating the detection of violations and integrating real-time notifications, the system addresses the inefficiencies of manual monitoring.

Key Achievements:

• Real-time processing of video feeds to calculate vehicle speeds.
• Accurate detection and recognition of license plates.
• Instant email notifications to authorities with violation evidence.

This project highlights the benefits of leveraging high-performance computing and AI in solving real-world challenges. With further enhancements, such as integration with city-wide traffic systems or extending capabilities to handle multiple video feeds simultaneously, the solution can significantly contribute to building smarter cities.

The complete codebase for the Intelligent Speed Monitoring and Alert System is available on GitHub. Visit the GitHub repository to explore the source code, annotated datasets, and implementation details.

GitHub Link

Thank you for taking the time to go through my project! I am truly grateful for the opportunity to participate in the global Computer Vision Projects Expo 2024 and share my work with such an inspiring community.