Superquadric Primitive Decomposition of 3D point clouds via Geometric-Aware Inlier Refinement

2026-09-28

AI Summary

This study addresses the susceptibility of superquadric decomposition of 3D point clouds to noise, outliers, and overlapping structures, which frequently leads to erroneous inlier misassignment. To overcome this limitation, we propose a geometry-aware framework that transcends conventional residual-based criteria by explicitly incorporating local geometric priors, such as normal consistency, into the fitting process. Furthermore, graph-cut optimization is employed to minimize an energy function for precise inlier refinement. This approach effectively suppresses the propagation of false inliers and stabilizes parameter estimation. Extensive evaluations on both synthetic and real-world datasets demonstrate that the proposed method significantly outperforms RANSAC in terms of geometric accuracy, robustness to noise, and convergence efficiency.

Abstract

The decomposition of 3D point clouds into interpretable geometric primitives remains a longstanding challenge in Computer Vision and Computer Graphics. Among the available representations, superquadrics offer a compact and expressive model capable of capturing a wide range of shapes. However, their estimation is inherently challenging, as it requires solving a non-linear optimization problem and is particularly sensitive to noise, outliers, and overlapping structures. While robust estimation methods such as RANSAC and its variants achieve strong performance, they rely primarily on spatial proximity and residual-based criteria, often leading to incorrect inlier assignments across adjacent or complex arrangements of primitives. In this work, we introduce a geometric-aware framework for primitive decomposition that explicitly incorporates local surface properties into the fitting process. Specifically, we propose an inlier refinement step formulated as an energy minimization problem and solved via graph-cut optimization. Our formulation integrates geometric priors, such as normal consistency, enabling more reliable inlier selection beyond purely residual-based criteria. The approach naturally applies to both single-model estimation and multi-model decomposition. By leveraging geometric information beyond point-wise residuals, our method reduces erroneous inlier propagation and stabilizes parameter estimation. Experiments on synthetic and real datasets show consistent improvements in geometric accuracy, robustness to noise and outliers, and convergence efficiency compared to state-of-the-art RANSAC-based methods.

Problem

• 3D point clouds
• superquadric primitive decomposition
• inlier assignment
• non-linear optimization
• geometric primitives

Innovation

• Superquadric Primitive Decomposition
• Geometric-Aware Inlier Refinement
• Graph-Cut Optimization
• Normal Consistency
• Energy Minimization