Enhancing Pose Estimation Stability and Accuracy for Fiducial Markers Using Transparent Cylinders
Hideyuki Tanaka and Kunihiro Ogata
AI summary
Problem
Conventional planar fiducial markers suffer from reduced orientation accuracy in near-frontal views, pose ambiguity, and limited depth accuracy, while previous high-accuracy alternatives require specialized, costly optical components.
Approach
The authors replace specialized lenses with simple transparent acrylic cylinders wrapped in printed black-and-white patterns to create cylindrical optical elements that continuously track orientation and detect pose inversion via geometric optics.
Key results
- Achieved a measurable orientation range of approximately -74° to +74°
- Reduced frontal-view orientation errors to ~0.4°
- Enabled robust pose-inversion detection and stable depth estimation
- Demonstrated low-cost, self-fabricatable marker construction using widely available materials
Why it matters
Provides an accessible, high-accuracy pose estimation tool for robotics and AR applications without relying on expensive or specialized manufacturing.
Abstract
Planar fiducial markers are widely used for 6-DoF pose estimation in robotics and augmented reality; however, conventional markers suffer from reduced orientation accu- racy in near-frontal views, pose ambiguity, and limited depth accuracy. We previously addressed these issues using lenticular- based high-accuracy markers, but their fabrication required specialized optical components. This paper proposes a novel self-fabricable fiducial marker that achieves comparable ac- curacy using transparent cylindrical rods. By attaching simple black-and-white printed patterns to acrylic cylinders, we realize two optical elements: a cylindrical Lenticular Angle Gauge (c- LEAG), which provides continuous orientation information, and a cylindrical Flip Detection Pattern (c-FDP), which enables robust pose-inversion detection. We analytically derive the relationship between viewing direction and pattern displace- ment using geometric optics and show that the orientation response is approximately linear within ±15◦. Based on these components, we develop c-LentiMark, integrating conventional feature-based pose estimation with cylinder-based pose correc- tion. Experiments using a precision rotation stage demonstrate a measurable orientation range of approximately -74◦to +74◦and a significant improvement in frontal-view accuracy, reducing orientation errors to about 0.4◦. The proposed marker achieves high accuracy comparable to previous lenticular designs while allowing low-cost fabrication from widely available materials.