Key-Axis-based Localization of Symmetry Axes in 3D Objects Utilizing Geometry and Texture

Area of Science:

• Computer Vision
• Robotics
• Geometric Modeling

Background:

• Pose estimation for objects with rotational symmetry often suffers from ambiguities.
• Manual determination of symmetry axes in current methods reduces accuracy and efficiency.

Purpose of the Study:

• To propose an automated method for determining symmetry axes in pose estimation.
• To reduce reliance on manual settings for symmetry axes, thereby improving accuracy.
• To enhance the efficiency of symmetry axis localization.

Main Methods:

• Determining symmetry axis orders and angles based on object type.
• Defining and localizing two key axes with highest orders.
• Generating three orthogonal axes from key axes to compute all symmetry axes.
• Utilizing the ADI metric for geometric symmetry and introducing ADI-C for texture symmetry.

Main Results:

• Achieved a 9.80% reduction in symmetry axis localization error.
• Improved pose estimation accuracy by 1.64%.
• Demonstrated performance on LM-O, HB, and the new DSRSTO dataset across diverse symmetric objects.

Conclusions:

• The key-axis-based approach offers a more efficient and accurate method for symmetry axis localization in pose estimation.
• Automating symmetry axis determination is crucial for robust object recognition in computer vision.
• The proposed metrics and dataset facilitate further research in symmetric object pose estimation.