👉 中文版

SmartifCane

An open-source AI-powered smart cane that fuses multi-modal perception with graph-gated agents to assist the visually impaired.

Table of Contents

1. Project Overview
2. Key Features
3. System Architecture
4. Hardware Components
5. Quick Start
6. Directory Layout
7. Models & Data
8. Demo
9. Performance Benchmarks
10. Roadmap
11. Contributing
12. Community & Support
13. License
14. Acknowledgements

---

Project Overview

SmartifCane combines a lightweight visual-detection engine, graph-neural gates, and large-language-model prompts to deliver fine-grained spatial awareness and natural language feedback for outdoor and indoor navigation.

Key Features

Module	Keywords	Description
Visual Detection	Lightweight OD engine	Real-time obstacle & signage recognition
Semantic Voice	MoE Prompts ＋ LLM	Personalized, multi-lingual scene speech
Emotional Aid	Emotion + Memory Bank	Tone adaption & long-term user memory
Offline Mode	On-device tiny models	Core functions without Internet access

System Architecture

Camera → Detection Engine → MSDN ROI → GNN Gate → Prompt Template
                ↘               ↘
       Mic/Audio → ASR  →  LLM (DeepSeek)