Designing a Simple Fiducial Marker for Localization in Spatial Scenes Using Neural Networks

Area of Science:

• Computer Vision
• Augmented Reality
• Robotics

Background:

• Existing fiducial markers often encode information and have color constraints.
• Robust spatial localization is crucial for augmented reality (AR) applications.
• Imperfections and motion blur can degrade marker detection accuracy.

Purpose of the Study:

• To design a novel, information-free fiducial marker for AR applications.
• To develop a marker system resistant to motion blur and imperfections.
• To enable precise spatial localization using any saturated color.

Main Methods:

• Designed a simple fiducial marker: a solid color rectangle with three lines from the center to corners.
• Utilized neural networks for precise marker detection with varied training data.
• Tested the marker's performance in diverse indoor and outdoor conditions.

Main Results:

• The marker accurately provides position, rotation, relative size, and projective transformation.
• The system demonstrates resilience to motion blur and marker imperfections.
• Detection accuracy was confirmed in both indoor and outdoor environments.

Conclusions:

• The proposed fiducial marker design is novel and effective for spatial localization in AR.
• The marker's flexibility in color choice enhances environmental blending.
• The system offers a robust and precise solution for AR spatial awareness.