CylinderTag: An Accurate and Flexible Marker for Cylinder-Shape Objects Pose Estimation Based on Projective...

Area of Science:

• Computer Vision
• Robotics
• Geometric Computer Vision

Background:

• High-precision pose estimation using visual markers is crucial in computer vision.
• Traditional flat markers are unsuitable for curved surfaces, limiting pose estimation accuracy.
• Developing robust markers for curved objects remains a significant challenge.

Purpose of the Study:

• To propose a novel visual marker, CylinderTag, specifically designed for pose estimation on developable curved surfaces, particularly cylindrical ones.
• To introduce a marker generator and recognizer to enhance CylinderTag's practical usability.
• To comprehensively evaluate CylinderTag's performance characteristics.

Main Methods:

• CylinderTag utilizes a cyclic design for secure attachment to cylindrical objects.
• It leverages manifold assumption and cross-ratio projective invariance for encoding.
• A heuristic search-based generator and a high-performance recognizer were developed.

Main Results:

• CylinderTag demonstrates superior detection performance across various view angles compared to traditional markers.
• It achieves higher localization accuracy and exhibits real-time detection capabilities.
• Extensive experiments confirm its robust performance in terms of detection rate, speed, and accuracy.

Conclusions:

• CylinderTag is a highly effective and promising visual marker for high-precision pose estimation on cylindrical surfaces.
• It offers enhanced versatility, practicality, and improved accuracy for curved object localization.
• The findings provide valuable guidance for future research in visual localization of cylinder-shaped objects.