dvinix/navora

🧭 Navora - AI Navigation Assistant for the Blind

Real-time assistive vision system providing navigation guidance for blind and visually impaired users.

✨ Features

AI-Powered Detection

• —YOLOv8 — Real-time object detection
• —BLIP-2 — Scene understanding and description
• —MiDaS — Depth estimation (GPU only)

Accessibility Features

• —🎯 Visual Feedback — Thin bounding boxes around detected objects
• —🗣️ Voice Guidance — Clear audio navigation instructions
• —📢 Scene Description — Automatic description when standing still
• —📳 Haptic Feedback — Vibration patterns for different obstacles
• —🚨 Smart Alerts — Context-aware, non-repetitive announcements

🚀 How to Use

On Mobile Device

1. 1.Open this Space on your mobile browser
2. 2.Allow camera access (uses back camera)
3. 3.Point camera at your path
4. 4.Receive real-time guidance:
5. 5.Voice: Spoken navigation instructions
6. 6.Vibration: Tactile feedback for obstacles
7. 7.Visual: Bounding boxes (for sighted assistants)

Navigation Actions

• —Forward ⬆️ — Path is clear, move forward
• —Stop 🛑 — Obstacle directly ahead
• —Left ⬅️ — Obstacle on right, move left
• —Right ➡️ — Obstacle on left, move right

Scene Description

• —Stand still for 5 seconds
• —System automatically describes surroundings
• —Example: "5 objects detected: 2 persons, 1 car, 1 bicycle"

🎯 Obstacle Detection

The system prioritizes obstacles based on:

• —Position — Center obstacles are highest priority
• —Confidence — Higher confidence = more reliable
• —Type — People, vehicles prioritized over static objects
• —Size — Larger objects get more attention

🔧 Technical Details

Architecture

Camera → YOLOv8 Detection → Priority Analysis → Navigation Guidance
                                                ↓
                                    Voice + Vibration + Visual

Performance

• —Inference: ~1-2 seconds per frame (CPU)
• —Frame Rate: ~1-2 FPS (optimized for battery)
• —Model Size: 6MB (YOLOv8n)
• —Accuracy: 45%+ confidence threshold

Optimizations

• —CPU-optimized model selection
• —Adaptive frame rate
• —Smart alert deduplication
• —Low battery mode support

📱 Mobile Compatibility

Works on:

• —✅ iOS Safari (iPhone/iPad)
• —✅ Android Chrome
• —✅ Android Firefox
• —⚠️ Requires HTTPS for camera access

🔒 Privacy

• —All processing happens on server
• —No video storage
• —No user tracking
• —Session data cleared after 2 minutes

🚧 Limitations

• —Requires internet connection
• —Works best in good lighting
• —May struggle with:
• —Very dark environments
• —Fast movement
• —Very crowded scenes
• —Reflective surfaces

🔮 Future Enhancements

• —📱 Mobile app with offline capability
• —🌙 Night vision mode
• —🗺️ Landmark recognition
• —🚨 Emergency SOS feature
• —🎤 Voice commands
• —🔊 Spatial audio (3D sound)

📚 Documentation

• —Accessibility Features
• —Implementation Guide
• —Mobile Edge Deployment

🤝 Contributing

Feedback from blind users is invaluable! Please test and share your experience.

📄 License

MIT License - Free to use and modify