Skip to content

nathan-cannell/AI_assisted_navigation_headgear

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Assistive Smart Device for Visually Impaired

Capstone Project – University of Washington ECE Department


Overview

This project presents a wearable assistive device designed to enhance navigation and environmental awareness for visually impaired users. Integrating real-time navigation and computer vision, the device provides spoken guidance and object detection through local processing, supporting independent and safe mobility in both indoor and outdoor environments.

Watch the Demo!

Watch a video of the device working in action at this youtube link:

(https://youtube.com/shorts/SWpah7KVuFU?feature=share)

IMG_1050 IMG_1048

Table of Contents


Team

Name Role
Kyshawn Savone-Warren Project Manager, Depth Sensing & Obstacle Avoidance
Carlos Alberto Morelos Escalera Object Detection, STM32 Hardware Integration
Luke Liu Computer Vision, AI Optimization
Santos Zaid STM32 Hardware Integration, System Communication
Nathan Cannell Navigation System, Audio Feedback

Project Goals

  • Enable safe navigation for visually impaired users via real-time GPS guidance and obstacle avoidance.
  • Detect and identify key objects (e.g., street signs, crosswalks, vehicles, pedestrians) using computer vision (YOLOv11).
  • Deliver real-time audio feedback for navigation and environmental awareness.
  • Operate as a standalone, wearable device with all computation performed locally for privacy and efficiency.

System Architecture

  • Wearable Helmet Design: Sensors and cameras are mounted on a helmet frame; processing units are housed on top.
  • Processing Units: STM32 microcontroller (sensor integration, audio output) and Raspberry Pi (object detection, navigation).
  • Sensor Suite: Arducam Time-of-Flight (ToF) camera for depth sensing, USB RGB camera for object detection, GPS module for navigation.
  • Audio Output: Speaker and amplifier for real-time voice feedback.
  • Power: Portable battery pack for all-day use.

Data Flow Diagram:

  • Cameras and GPS feed data to Raspberry Pi.
  • Raspberry Pi runs navigation and object detection algorithms.
  • STM32 handles sensor integration and audio output.
  • Communication between Raspberry Pi and STM32 via USB/UART.
  • Audio cues delivered to the user in real time.

Key Features

  • Obstacle Avoidance: ToF camera detects obstacles (2–4m range) and provides spatial alerts.
  • Navigation: GPS-based turn-by-turn guidance with spoken instructions.
  • Object Detection: YOLOv11 model identifies street signs, crosswalks, vehicles, and pedestrians.
  • Audio Feedback: Real-time speech output for navigation and object alerts.
  • Local Processing: All computation is performed on-device, ensuring privacy and low latency.
  • Indoor/Outdoor Use: Optimized for urban navigation and large indoor spaces.

Hardware Components

Component Function Cost
Arducam ToF Camera Depth/Obstacle Sensing $109.98
USB Camera Object Detection $15.99
Audio Amplifier Audio Output $5.19
Speaker Audio Output N/A*
GPS Module Navigation $12.99
Buck Converter Power Regulation $9.99
Power Supply Battery N/A*

*Items marked N/A were already available in lab supply.


Software Components

  • YOLOv11: Real-time object detection (trained on custom and public datasets).
  • OSMnx: Offline route planning and navigation.
  • Custom Depth Sensing Scripts: For obstacle detection and spatial cueing.
  • Audio Subsystem: Text-to-speech and audio output via STM32 and CS43L22 codec.
  • Inter-device Communication: USB/UART protocols for data exchange between Raspberry Pi and STM32.

Engineering Constraints & Standards

  • Processing Power: Embedded hardware (Raspberry Pi, STM32) with optimized models for real-time performance.
  • Power Consumption: Battery-powered for portability; power management is critical.
  • Environmental Factors: IP54 water/dust resistance; operates in varied lighting and weather.
  • Safety: Compliance with ISO 13482:2014 for assistive devices and ANSI S3.5-1997 for speech intelligibility.
  • Wireless Communication: IEEE 802.11 for debugging and logging during development.

Development & Testing

Subsystems:

  • Navigation: GPS integration, OSMnx route planning, turn-by-turn audio cues.
  • Object Avoidance: ToF camera integration, spatial analysis (left/right detection), real-time alerts.
  • Computer Vision: YOLOv11 training and optimization, real-time inference on Raspberry Pi.
  • Audio: STM32 audio output via CS43L22 codec, text-to-speech, latency testing.

Integration:

  • Synchronized data flow between navigation, object detection, and audio feedback.
  • Local processing; no cloud reliance for privacy.
  • Iterative testing in real-world urban and indoor environments, including low-light scenarios.

Experimental Outcomes

  • Object Detection: Achieved >70% accuracy for key street signs and objects in optimal conditions.
  • Obstacle Avoidance: Reliable detection within 2–4m; two-part spatial analysis (left/right) provided effective alerts.
  • Navigation: Turn-by-turn guidance functional; GPS accuracy limited indoors.
  • Audio Feedback: Real-time spoken cues with <500ms latency.
  • System Integration: All subsystems operated cohesively in prototype testing, with successful real-world trials.

Impact & Future Work

  • Social Impact: Promotes independence and inclusion for visually impaired individuals in urban and campus environments.
  • Commercial Potential: Viable for further development and commercialization, especially in collaboration with industry partners.
  • Future Improvements:
    • Enhance processing efficiency and battery life.
    • Refine AI models for improved accuracy in challenging conditions.
    • Integrate with existing navigation apps.
    • Conduct broader user testing and collect feedback for iterative design.

Acknowledgements & Licensing

Special thanks to Professor Hussein, the ECE Department, and the UW teaching staff for their support and funding.

Licensing:
This project uses open-source components (YOLOv11, OSMnx, Arducam ToF API) under their respective licenses. All original hardware, software, and documentation are © the project team. For any use or reproduction, proper attribution is required.


Contact:
For questions or collaboration, please contact the project team via the UW ECE Department.


About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages