Blind-AI: Smart Assistive Navigation Device for Visually Impaired

This project implements a smart assistive device, integrated into a wearable hat, designed to enhance independent navigation for visually impaired individuals. It combines real-time object detection, distance measurement, GPS-based navigation, and voice interaction to provide comprehensive environmental awareness and guidance, all within a discreet and convenient form factor.

Overview

The system, seamlessly housed within a hat, consists of a client device (running on a Raspberry Pi) and a remote server.

Client (in client/)

The client-side application captures real-time video and interacts with the user through voice commands. It features two primary modes:

• Guidance Mode:
  • Captures video frames from a camera.
  • Sends video frames to the remote server for object detection.
  • Receives and provides audio feedback on detected objects.
  • Uses an ultrasonic sensor to detect nearby obstacles and alerts the user with beeps.
• Chat Mode:
  • Utilizes OpenAI to engage in conversational interactions based on voice input.

The client listens for a wake word ("Hi siri") to activate and can be shut down with a "bye" command. It employs threading for concurrent operation and includes error handling and resource management.

Server (in server/)

The server-side application processes video frames for object detection and integrates this information with GPS navigation updates.

• main.py (Server Main):

  • Sets up a server to receive video frames from multiple clients.
  • Utilizes an object detection model (YOLOv8) to identify objects in the video streams.
  • Combines object detection results with GPS navigation updates (provided by the Flask server).
  • Sends the combined information back to the client as text.
  • Manages multiple clients concurrently using threads.
  • Runs a background Flask server (gps_handler.py) to handle GPS data.

• predictions.py (Object Detection):

  • Contains functions for processing video frames using a YOLO model to detect specific objects.
  • Identifies the closest detected object to the center of the frame.
  • Determines the object's relative position (left, right, or ahead).
  • Generates directional instructions for the client.
  • Implements logic to avoid repetitive reporting of the same object based on time and a delay.

• gps_handler.py (GPS Navigation Server):

  • Creates a Flask web server to provide GPS-based turn-by-turn navigation instructions.
  • Receives the user's current latitude and longitude via POST requests to the /location endpoint (likely from a mobile application).
  • Uses the Google Maps Directions API to calculate a route to a predefined destination.
  • Determines the next navigation instruction based on the user's current location and the route steps, considering distance to turning points or the destination.
  • Formats navigation instructions for clarity.
  • Uses a thread lock for safe access to location data.

Network Setup

The Raspberry Pi (client) and the server need to be on the same network or accessible via a public IP address. This project was developed and tested using Tailscale VPN, which creates a secure private network allowing seamless communication between the mobile app, client, and server, regardless of their physical location.

Functionality

This framework aims to provide visually impaired individuals with:

• Real-time obstacle detection: Identifying and alerting the user to nearby obstacles.
• Object recognition: Providing information about objects in the environment.
• Turn-by-turn GPS navigation: Guiding the user to their destination.
• Voice interaction: Enabling intuitive communication with the device for commands and information.

Project Execution Requirements

Hardware Requirements:

• Raspberry Pi 5 (8GB recommended)
• Raspberry Pi Camera Module or compatible USB webcam
• Ultrasonic Sensor (HC-SR04) – for obstacle distance detection
• Buzzer – for proximity alerts
• Microphone Module – for voice command input
• Speaker / Earphones – for audio feedback (Google TTS output)
• Power Supply – Portable power bank (minimum 10,000 mAh)
• Wi-Fi connectivity – Required for Raspberry Pi to communicate with server
• Smartphone – To run GPS tracking app and send coordinates
• Optional: Breadboard, jumper wires, resistor (for ultrasonic circuit setup)

Software Requirements:

On Raspberry Pi (Client Side):

• OS: Raspberry Pi OS (32-bit recommended)
• Python Version: Python 3.8 or later
• Required Python Libraries: pip install SpeechRecognition gtts playsound requests socket numpy pyaudio RPi.GPIO
• Google APIs:
  • Google Speech-to-Text API
  • Google Text-to-Speech (gTTS)
• Flask Client (to interact with server)
• Mic & audio configuration (ALSA or PulseAudio)

On Server (Object Detection & Navigation Logic):

• OS: Windows/Linux (recommended: Ubuntu or any local machine)
• Python Version: Python 3.8 or later
• Required Python Libraries: pip install flask ultralytics opencv-python numpy requests geopy
• YOLOv8 Model Files:
  • Install Ultralytics: pip install ultralytics
  • Use pre-trained model: yolov8n.pt or yolov8s.pt

On Mobile (GPS Tracking App):

• Mobile app
• Send location data to the server via REST API POST (Flask route)
• Smartphone must have:
  • GPS turned ON
  • Mobile data/Wi-Fi enabled

Other Essentials:

• Raspberry Pi and server must be on the same network or accessible via public IP.
• Update IP addresses in the socket/client/server scripts as per your setup.

Authors

• Joe Mathew
• Joel Kurian Abraham
• Nived Tharun
• Muhammed Jasim PT