Low-cost Computer Vision System for Autonomous Outdoor Robots

This project formulates my final year dissertation project for my BSc in Computer Science. It is a computer vision-based navigation system using ArUco fiducial markers for guiding autonomous robots through gate formations.

Overview

This project implements a low-cost vision system using ArUco markers and a Raspberry Pi to provide directional guidance for autonomous outdoor robots. The system:

• Uses custom ArUco markers arranged in gates to create a navigation path
• Detects markers using computer vision (OpenCV)
• Calculates target points between marker pairs
• Provides directional output via GPIO pins or UART serial communication
• Runs on resource-constrained hardware (Raspberry Pi Zero)

Components

The system uses:

ArUco Markers

• Custom 4x4 bit ArUco markers (4 types):
  • Left marker (ID 0)
  • Right marker (ID 1)
  • Start marker (ID 2)
  • Stop marker (ID 3)

Hardware

• Raspberry Pi Zero with camera
• OpenCV for computer vision
• Serial UART communication

Installation

Install OpenCV and dependencies

pip install opencv-python
pip install numpy
pip install pyserial

Usage

Running on Desktop/Development Machine

Test with image input

python src/detectarucoimage.py

Test with video input from camera

python src/detectarucovideo.py [camera_index]

Running on Raspberry Pi

Main navigation program

python src/raspiaruco.py [camera_index]

Test serial communication

 python src/testserial.py  # Send test message
 python src/readserial.py  # Read serial output

Gate Formation

• Place left and right ArUco markers in parallel to form gates
• Start marker initiates navigation
• Stop marker ends navigation
• System calculates target point between valid marker pairs
• Direction is encoded as ASCII characters (A-Z) based on target position

How It Works

1. Camera captures video feed
2. OpenCV detects ArUco markers in each frame
3. System identifies valid gate formations:
   • Left marker must be on left side
   • Right marker must be on right side
   • Markers should be roughly parallel and similar size
4. Target point calculated between valid marker pairs
5. Target x-coordinate scaled to A-Z ASCII character
6. Direction sent via UART/GPIO to robot controller

Project Structure

 ├── src/
 │   ├── detectarucoimage.py   # Image detection test
 │   ├── detectarucovideo.py   # Video detection test  
 │   ├── raspiaruco.py         # Main Raspberry Pi program
 │   ├── readserial.py         # Serial read test
 │   └── testserial.py         # Serial write test
 ├── Images/                   # Documentation images
 └── README.md

Author

Sharan Govinden Umavassee

Acknowledgments

• Project Supervisor: Dr. Klaus-Peter Zauner
• Second Examiner: Prof. Christopher Freeman
• University of Southampton, School of Electronics and Computer Science

License

This project is part of MEng Computer Science with Industrial Studies at University of Southampton.