Mobility aid

Huia Echo 🐸

👁️‍🗨️ Mobility Aid for Visually Impaired Users

A low-cost embedded system designed to assist visually impaired individuals using real-time proximity sensing and traffic light recognition. Built on a Raspberry Pi, the system integrates ultrasonic sensors, a camera, and a lightweight machine learning model (YOLOv11n) to detect environmental conditions and provide haptic feedback.

📌 Overview

This mobility aid project helps users safely navigate streets, particularly at pedestrian crossings. The system combines:

• Ultrasonic sensing to detect nearby obstacles
• Traffic light recognition using a camera and an efficient YOLOv11n model
• Feedback mechanism via vibration, triggered based on proximity and light state

📁 Key File Structure

mobility-aid/
│
├── include/                  # C header files
│   ├── config_loader.h       # Loads external config values
│   ├── gpio_functions.h      # GPIO pin setup and control
│   └── tick.h                # Timing utilities
│
├── src/                      # C source files
│   ├── configs/              # Config database using data generated by web app
│   │   └── config.db
│   └── sys_controller.c      # Ultrasonic sensor and control loop                         
│
├── ml/                       # Machine learning module
│   ├── augmentation.ipynb    # Data augmentation and preprocessing
│   ├── train_model.ipynb     # YOLOv11n training logic
│   └── models/
│       └── best.pt           # Trained YOLOv11n model weights
│
├── test/                     # Test cases, mocks and framework files for testing C & Python
│
├── web/                      # Web interface (Flask app)
│   └── app.py                # Local dashboard
│
├── Makefile                  # Compiles the C code

🚀 Features

• 📡 Ultrasonic Sensing
  Detects objects and measures distance with HC-SR04 sensor.

• 🧠 Traffic Light Recognition (YOLOv11n)
  Recognizes red/green pedestrian and traffic lights from a camera feed.

• 🔊 Feedback System
  Triggers vibration when objects are too close or crossing is unsafe.

• ⚙️ Modular Design
  Split into embedded C for performance and Python for AI tasks.

• 🪛 Easily Configurable
  Adjust logic thresholds via external config files.

🧠 Machine Learning

• Model: YOLOv11n (Ultralytics)
• Dataset: Custom images gathered and augmented to increase dataset size
• Use Case: Detect green/red pedestrian signals in real time
• Output: Torch .pt model located at ml/models/best.pt
• Trained Using:
  • augmentation.ipynb: Image resizing, color shifts, flips
  • train_model.ipynb: Fine-tuning YOLOv11n with labeled traffic light data

🖥️ Requirements

Hardware

• Raspberry Pi 4 (or newer)
• HC-SR04 Ultrasonic sensor
• Pi Camera Module
• Vibration motor or buzzer