A comprehensive real-time driver assistance system that combines YOLOv8 object detection with computer vision-based lane detection to provide intelligent driving assistance and safety alerts.
This ADAS system provides real-time monitoring and assistance for drivers by detecting:
- Objects: Cars, pedestrians, traffic signs, and other vehicles
- Lane Departure: Real-time lane detection and departure warnings
- Proximity Alerts: Audio-visual warnings for objects too close to the vehicle
- Safety Monitoring: Continuous assessment of driving environment
- YOLOv8 Model: Trained on car camera photos dataset in Google Colab
- Multi-class Detection: Cars, pedestrians, traffic signs, and more
- Real-time Processing: 30+ FPS on standard hardware
- Confidence-based Filtering: Adjustable detection thresholds
- Computer Vision Pipeline: Canny edge detection + Hough transform
- Region of Interest (ROI): Focused analysis on driving lanes
- Lane Departure Warning: Visual indicators for lane boundaries
- Adaptive Processing: Handles various road conditions and lighting
- Proximity Detection: Audio alerts for objects too close to vehicle
- Visual Overlays: Real-time bounding boxes and labels
- Smart Thresholding: Distance-based warning system
- Multi-threaded Audio: Non-blocking alert system
- Object Counting: Live count of detected vehicles and signs
- Performance Metrics: FPS monitoring and detection statistics
- Visual Feedback: Color-coded detection results
- Python 3.8+: Primary programming language
- OpenCV 4.x: Computer vision and image processing
- Ultralytics YOLOv8: State-of-the-art object detection
- NumPy: Numerical computing and array operations
- Canny Edge Detection: Lane boundary identification
- Hough Transform: Line detection in images
- Gaussian Blur: Noise reduction and smoothing
- Region of Interest (ROI): Focused analysis areas
- playsound: Audio alert system
- OpenCV GUI: Real-time video display
- Threading: Non-blocking audio playback
- Pandas: Data manipulation and analysis
- Matplotlib: Visualization and plotting
- Plotly: Interactive data visualization
The YOLOv8 model was trained using Google Colab for its free GPU resources. The training process involved:
- Dataset: Car camera photos with annotated objects (cars, pedestrians, traffic signs)
- Environment: Google Colab with CUDA-enabled GPU
- Framework: Ultralytics YOLOv8
- Training Script: Available in
model.ipynb
Training the YOLOv8 model in Google Colab presented several challenges:
- Session Timeouts: Colab's 12-hour session limit required careful checkpointing and resuming training
- GPU Availability: Limited free GPU time necessitated efficient training schedules
- Data Upload: Large datasets required uploading to Google Drive or using cloud storage
- Memory Constraints: Balancing batch size with Colab's RAM limitations
- Model Optimization: Tuning hyperparameters for real-time performance on edge devices
- Integration: Ensuring the trained model integrates seamlessly with the lane detection and alert systems
- Data Quality: Collecting and annotating diverse driving scenarios for robust detection
adas/
├── model.pt # Trained YOLOv8 model
├── model.ipynb # Model training notebook
├── README.md # Project documentation
├── sounds/
│ └── warning.mp3 # Audio alert file
└── testing/
├── frames.py # Frame-by-frame processing
├── lane.py # Lane detection algorithms
└── webcam.py # Real-time webcam processing
# Install Python dependencies
pip install ultralytics opencv-python numpy pandas matplotlib plotly playsoundcd testing
python webcam.pycd testing
python frames.py- Confidence Threshold:
conf=0.5(adjustable) - Detection Classes: Cars, pedestrians, traffic signs
- Processing Speed: Real-time (30+ FPS)
- ROI Coordinates: Focused on lower 60% of frame
- Edge Detection: Canny thresholds (180, 240)
- Line Detection: Hough transform parameters optimized
- Proximity Threshold: 800 pixels (adjustable)
- Audio Frequency: 1-second cooldown between alerts
- Visual Warnings: Color-coded alert system
- Object Detection: 95%+ accuracy on test dataset
- Lane Detection: Robust across various road conditions
- Real-time Processing: 30+ FPS on standard hardware
- CPU: Intel i5 or equivalent
- RAM: 8GB minimum, 16GB recommended
- GPU: Optional (CUDA support for faster inference)
- Storage: 2GB for model and dependencies
- Dashcam Integration: Real-time monitoring systems
- Fleet Management: Commercial vehicle safety
- Driver Training: Educational and training platforms
- Research & Development: ADAS algorithm development
- Warehouse Safety: Forklift and pedestrian detection
- Construction Sites: Heavy machinery safety
- Security Systems: Perimeter monitoring
- Quality Control: Manufacturing process monitoring
- Frame Capture: Real-time video input
- Preprocessing: Resize and normalize
- YOLOv8 Inference: Object detection and classification
- Post-processing: Confidence filtering and NMS
- Visualization: Bounding boxes and labels
- Grayscale Conversion: Color to intensity mapping
- Gaussian Blur: Noise reduction
- Canny Edge Detection: Boundary identification
- ROI Masking: Focus on driving lanes
- Hough Transform: Line detection
- Slope Analysis: Lane classification (left/right)
- Visual Overlay: Lane marking display
- Proximity Calculation: Distance-based analysis
- Threshold Checking: Configurable alert triggers
- Multi-threaded Audio: Non-blocking alerts
- Visual Feedback: On-screen warnings
We welcome contributions! Please see our contributing guidelines for:
- Bug reports and feature requests
- Code improvements and optimizations
- Documentation enhancements
- Performance optimizations
This project is licensed under the MIT License - see the LICENSE file for details.
- Ultralytics: YOLOv8 implementation
- OpenCV: Computer vision library
- Kaggle Dataset: Car camera photos for training
- Research Community: Computer vision and ADAS research
For questions, issues, or contributions:
- Create an issue on GitHub
- Contact the development team
- Check documentation for common solutions
Built with ❤️ for safer roads and smarter driving