ai-powered-dashcam

🛣️ ADAS Same-Lane Proximity Filtering

The system only warns about objects inside your driving lane using a dynamic trapezoidal ROI — eliminating false alerts from vehicles on adjacent lanes.

          ┌──────┐          ← Lane Top (16% frame width)
         /        \
        /          \
       /            \
      └──────────────┘     ← Lane Bottom (70% frame width)

Proximity Levels (based on bounding box height ratio):

🚨 Alert System

Voice Alerts (pyttsx3 — spoken audio):

• "Animal close to the road" — animal in lane proximity
• "Pedestrian very close" — pedestrian in danger/warning zone
• "Camera view is blocked" — 80%+ of frame is occluded
• "Low visibility, drive carefully" — night/low light detected
• "Possible accident detected" — accident scoring threshold exceeded

Audio Beeps (pygame — synthesized tones):

🏎️ Speed Limit Recognition

A background thread submits road frames to a Roboflow-hosted Speed Limit Detection model every 5 seconds — detecting speed limit signs from live video and displaying them on the HUD in real-time.

💥 Accident Detection Engine (AccidentScorer)

Multi-signal scoring system that raises an alert when any combination of these triggers is detected:

• Frame Change Intensity: High pixel difference (cv2.absdiff) between consecutive frames
• Sustained Scene Change: Scene change above threshold for 3+ consecutive frames
• Object Count Drop: 2+ objects suddenly disappearing from frame (e.g. vehicles passing out of view at impact)

🌙 Environmental Awareness

📟 Live HUD Overlay

Real-time heads-up display rendered on the video feed:

• FPS counter — live frame rate
• Lighting condition — Day / Low Light / Night
• Speed Limit sign — when detected by Roboflow
• Status bar — COLLISION RISK / CLOSE PEDESTRIAN / CLOSE VEHICLE / CAMERA BLOCKED

📁 Project Structure

ai-powered-dashcam/
│
├── driverview.py     # Driver-facing camera system (USB Webcam / Index 1)
├── roadview.py       # Road-facing camera system (Laptop Camera / Index 0)
└── README.md

Note: Both scripts use the same integrated ADAS engine. Run them simultaneously for the full dual-camera experience — driverview.py on the interior camera and roadview.py on the forward-facing road camera.

🛠️ Installation

Prerequisites

• Python 3.8+
• Webcam(s) connected

1. Clone the Repository

git clone https://github.com/ashwinsiva007/ai-powered-dashcam.git
cd ai-powered-dashcam

2. Install Dependencies

pip install ultralytics opencv-python pyttsx3 pygame numpy roboflow

3. YOLOv8 Model

The yolov8n.pt model is automatically downloaded on first run via Ultralytics. No manual download needed.

🚀 Usage

Run the Road Camera (Front View)

python roadview.py

Run the Driver Camera (Interior View)

python driverview.py

Use a Video File Instead of Webcam

python roadview.py --source path/to/video.mp4

Dual-Camera Mode (Two Terminals)

# Terminal 1 — Road view
python roadview.py

# Terminal 2 — Driver view
python driverview.py

When prompted:

• Press Enter → uses connected USB webcam (Index 1) or falls back to laptop camera (Index 0)
• Enter a video file path → processes that video file

Press Q to quit either window.

📦 Dependencies

⚙️ Configuration

All thresholds and settings are centralized in the CONFIG dictionary at the top of each script:

CONFIG = {
    "danger_threshold":  0.55,   # bbox height ratio for danger
    "warning_threshold": 0.35,   # bbox height ratio for warning
    "caution_threshold": 0.18,   # bbox height ratio for caution
    "beep_cooldown":     1.5,    # seconds between audio beeps
    "confidence":        0.40,   # YOLO detection confidence
    "frame_skip":        2,      # process every N frames (performance)
    "lane_top_width":    0.16,   # lane ROI top width (% of frame)
    "lane_bottom_width": 0.70,   # lane ROI bottom width (% of frame)
}

🔮 Roadmap

• Driver drowsiness detection (eye aspect ratio / facial landmarks)
• SOS auto-alert integration (SMS / emergency call on accident)
• Lane departure warning
• Mobile app integration (Flutter client)
• Edge deployment (Raspberry Pi / Jetson Nano)
• Cloud dashboard for trip recording & replay

Built with ❤️ by Ashwin Sivaram

⭐ Star this repo if you find it useful!