Compact World Anchors: Registration Using Parametric Primitives as Scene Description.

Area of Science:

• Computer Vision
• Robotics
• Geometric Modeling

Background:

• Accurate 3D registration is crucial for augmented reality (AR) and robotics.
• Existing methods often struggle with efficiency and robustness in real-world scenarios.
• Leveraging parametric models offers a promising avenue for improved registration.

Purpose of the Study:

• To develop a closed-form registration method using geometric constraints from parametric primitives.
• To enable robust scene-to-model correspondence matching for 3D data.
• To facilitate compact world anchor creation for indoor localization in AR applications.

Main Methods:

• Extracting geometric constraints (lines, planes) from parametric primitives in 3D models.
• Utilizing semantically segmented RGB-D scans or plane detection outputs (ARKit, ARCore).
• Building and matching primitive descriptors (normals, centroids) for scene-to-model correspondence.
• Employing a closed-form solver for estimating 6-DOF transformation from line and point correspondences.

Main Results:

• Demonstrated performance and robustness on synthetic and real-world datasets.
• Successful creation of compact world anchors for indoor localization.
• Effective registration using line-to-line, line-to-plane, and plane-to-plane correspondences.
• Validation of the method's applicability in AR on mobile devices leveraging SLAM.

Conclusions:

• The proposed method provides an efficient and robust solution for 3D registration.
• Parametric primitive-based constraints offer a powerful approach for scene-to-model matching.
• The technique is well-suited for enabling accurate indoor localization in mobile AR applications.