MAI642 Team DeepWave: Vision-Based Parking Management System Using Optimized YOLOv11
Project Overview
This project presents an innovative parking management solution using advanced computer vision and deep learning techniques. The system aims to modernize parking management by providing accurate, real-time information about parking space availability.
Problem Statement
Traditional parking systems often face challenges such as:
- Difficulty in finding available parking spaces
- Inaccurate availability information
- Long waiting times for parking
Mission
Our mission is to:
- Modernize and enhance parking management systems
- Improve customer experience
- Provide precise and accurate parking space information
Key Features
- Real-time parking space detection
- Vehicle occupancy tracking
- Optimized YOLO object detection model
- Drone-based video monitoring
Technical Approach
Model Development
- Base Model: YOLOv11
- Backbone: Custom EfficientNet integration
- Key Modifications:
- Replaced original backbone with EfficientNet
- Created custom configuration file (yolo11_EfficientNet.yaml)
- Implemented core EfficientNet classes and modules
Dataset
- Source: https://universe.roboflow.com/ucy-dlyme/mai642_deep_learning-deepwave
- Data Split:
- 70% Training
- 20% Validation
- 10% Testing
- Data Collection: Over 5000 images
- Data Augmentation Techniques:
- Image flipping
- Rotation
- Noise addition
Performance Metrics
Model | Precision | Recall | MAP50 | MAP50-95 |
---|---|---|---|---|
YOLOv11s | 0.958 | 0.933 | 0.971 | 0.757 |
YOLOv11s (frozen layers) | 0.918 | 0.956 | 0.974 | 0.758 |
YOLOv11n (frozen layers) | 0.959 | 0.902 | 0.902 | 0.717 |
Expected Benefits
- 35% Reduction in customer waiting times
- 30% Reduction in operational costs
- 23% Increase in customer satisfaction
Project Workflow
- Data Collection and Preparation
- Model Training and Evaluation
- Model Configuration
- Testing and Workflow Optimization
- Deployment
Team Members
- Jianlin Ye: Dataset Creation, UAV Video Recording, YOLOv11 Backbone Replacement
- Rafael Koullouros: Dataset Creation, Model Training, Evaluation
- Kyriakos Pelekanos: Workflow Optimization
- Mikhail Sumskoi: HuggingFace Deployment, Basic UI
Repository
- GitHub: https://github.com/JYe9/YOLO11_EfficientNet
- HuggingFace: https://huggingface.co/jye9/DeepWave
- Dataset: https://universe.roboflow.com/ucy-dlyme/mai642_deep_learning-deepwave
Deployment
- Platform: HuggingFace (for demonstration)
Future Work
- Expand dataset
- Further optimize model performance
- Develop more comprehensive UI
- Implement wider parking management features
Model tree for jye9/DeepWave
Base model
Ultralytics/YOLO11