AccessAI Navigation Assistant

Real-time navigation assistance for visually-impaired users using an Intel RealSense D435i depth camera and YOLOv8 object detection.

Features

• Object Detection: Identifies people, furniture, vehicles, and other obstacles
• Distance Estimation: Uses depth camera to measure how far away objects are
• Bump/Drop-off Detection: Warns about curbs, steps, and potholes
• Voice Warnings: Speaks alerts for nearby hazards (prioritizes closest dangers)
• Visual Display: Shows camera feed with color-coded bounding boxes

Requirements

• Python 3.9+
• Intel RealSense D435i camera (connected via USB 3.0)
• Windows 10/11

Installation

1. Create a virtual environment (recommended):

   python -m venv venv
   venv\Scripts\activate
   

2. Install dependencies:

   pip install pyrealsense2 opencv-python numpy ultralytics pyttsx3 pillow
   

Running the Application

1. Connect your Intel RealSense D435i camera via USB 3.0

2. Navigate to the project folder:

   cd AccessAIThon
   

3. Run the application:

   python main.py
   

4. Click "Start Camera" to begin detection

5. Click "Stop" or close the window to exit

Troubleshooting

"Error: Couldn't resolve requests"

• Make sure the camera is connected via USB 3.0 (not 2.0)
• Close any other apps using the camera (Intel RealSense Viewer, etc.)
• Try unplugging and reconnecting the camera

No audio warnings

• Check your system volume
• Make sure no other app is blocking audio

Low FPS

• The app runs YOLO every 5th frame for performance
• Close other heavy applications
• Ensure you're using the yolov8n.pt (nano) model

Configuration

Edit the constants at the top of main.py to adjust:

Setting	Default	Description
DANGER_DISTANCE_M	1.5	Distance (m) to trigger caution warning
CRITICAL_DISTANCE_M	0.6	Distance (m) to trigger urgent warning
AUDIO_COOLDOWN_S	2.0	Seconds between repeated warnings
PERSISTENCE_THRESHOLD	5	Frames needed to confirm a detection