How itsacchu/ai-ar-navigation works
Project type and purpose
• This is a self-contained AR navigation system for visually impaired users that runs on edge devices (mobile/embedded).
• It fuses video, audio, and inertial sensors; performs real-time AI inference; and outputs audio and haptic cues.
File structure
• main.py
– Entry point that orchestrates module initialization and the processing loop.
• depth_estimator.py
– Wraps the MiDaS model for per-frame distance estimation.
• object_detector.py
– Hosts YOLOv8/EfficientDet inference and traffic-sign OCR via EasyOCR.
• navigation_system.py
– Implements A* path planning, dynamic recalculation with OSM API, and safety buffering.
• requirements.txt / .gitignore / README.md
– Define dependencies and high-level architecture overview.
README for additional insights
• Four core functional modules:
1. Input Processing (video stabilization, ASR, Kalman-based sensor fusion)
2. Perception (object detection, depth estimation, path segmentation)
3. Navigation (A* planning, geospatial processing, dynamic updates)
4. User Interface (TTS via ElevenLabs, haptic feedback)
• Latency and accuracy targets (<150 ms, >90% object detection, <3 m deviation).