The Advanced Driver Assistance System (ADAS)

A production-ready Advanced Driver Assistance System (ADAS) powered by computer vision and deep learning. Trained on ~160 GB of driving datasets, this system integrates 8 safety-critical components, from lane departure warnings to driver drowsiness monitoring, delivering real-time performance with fully local processing.

Highlights

• 8 Production-Ready ADAS Components (Lane Detection, FCW, Pedestrian Detection, etc.)
• A Unified "The ADAS System" Mode, for running 7 road-based systems simultaneously
• Apple Silicon-optimized performance with MPS (Metal Performance Shaders)
• Fully Local Processing, no cloud dependencies after setup
• Interactive Streamlit UI with real-time running
• ~160 GB Training Data across multiple benchmark datasets

Key Features

8 Complete ADAS Components

1. Lane Detection + Lane Departure Warning (LDW)

• Model: UNet + ResNet18 Segmentation
• Dataset: CULane (~93 GB)
• Features:
  • Real-time lane boundary detection
  • Vehicle offset calculation from lane center
  • Configurable departure threshold warning

2. Forward Collision Warning (FCW)

• Model: YOLOv8n (COCO 2017)
• Dataset: COCO 2017 (~25 GB)
• Features:
  • Vehicle detection (cars, trucks, buses)
  • SORT multi-object tracking
  • Time-To-Collision (TTC) calculation with configurable threshold

3. Pedestrian Detection

• Model: YOLOv8n (COCO 2017)
• Dataset: COCO 2017 (~25 GB)
• Features:
  • Real-time pedestrian detection
  • Distance-based warning system
  • Height-based proximity estimation

4. Two-Wheeler Detection

• Model: YOLOv8n (COCO 2017)
• Dataset: COCO 2017 (~25 GB)
• Features:
  • Detects bicycles and motorcycles
  • Separate alerts for different vehicle types

5. Animal Awareness

• Model: YOLOv8n (COCO 2017)
• Dataset: COCO 2017 (~25 GB)
• Features:
  • Detects 6 animal classes (cat, dog, horse, sheep, cow, bear)
  • Distance-based warning system

6. Traffic Sign Recognition

• Model: YOLOv8n (MTSD)
• Dataset: MTSD (~8.1 GB)
• Features:
  • Detects common traffic sign types
  • Real-time sign classification
  • Configurable confidence thresholds

7. Traffic Light Detection

• Model: YOLOv8n (LISA)
• Dataset: LISA (~10 GB)
• Features:
  • Detects traffic lights in various conditions
  • Classifies state (Red, Yellow, Green, Off)

8. Driver Drowsiness Monitor

• Model: MediaPipe Face Mesh + EAR Algorithm
• Calibration: UTA-RLDD (~23 GB)
• Features:
  • 468-point facial landmark detection
  • Eye Aspect Ratio (EAR) monitoring
  • Fatigue and drowsiness alerts
  • Real-time blink rate analysis

The ADAS System (Combined Mode)

Run 7 road-based systems simultaneously:

• Lane Detection + LDW
• Forward Collision Warning
• Pedestrian Detection
• Two-Wheeler Detection
• Animal Awareness
• Traffic Sign Recognition
• Traffic Light Detection

Features:

• Consolidated multi-system visualization
• Unified warning aggregation
• Independent component enable/disable via UI
• Optimized for real-time multi-tasking

How To Run

1. Make sure you have Python 3.8+ set up, clone this repository on your local machine, and set up the required datasets.
2. Create a virtual environment, install the required dependencies and run the app:

pip install -r requirements.txt
streamlit run app.py

Contributing

Contributions are welcome!

License

Distributed under the MIT License.