ransac plane fitting

Traditional RANSAC-based plane detection in depth images suffers from erroneous detections due to ambiguous inlier thresholds, especially when multiple planar surfaces coexist and the true number of planes is unknown. To address this, we propose a model-information-optimized generic plane detection framework. Our method integrates the physical characteristics and noise model of depth sensors, and employs information-theoretic minimization to automatically determine the number of planes, estimate their parameters, and rank them by reliability. Crucially, we replace heuristic inlier thresholds with model information content as the core evaluation metric. Furthermore, we enhance computational efficiency via randomized subsampling for candidate model generation and neural-network-guided region segmentation. Experiments demonstrate that our approach outperforms Open3D’s default RANSAC on synthetic data, yielding more accurate parameter estimates; on real-world scenes, it produces significantly more robust and geometrically plausible planar segmentations.

To address challenges in architectural point cloud modeling—including severe noise interference, complex planar structures, and low semantic segmentation accuracy—this paper proposes an end-to-end point cloud processing framework. Methodologically, it integrates adaptive Z-score outlier removal, hierarchical RANSAC-based multi-plane detection, and a lightweight modified PointNet for semantic segmentation, forming a multi-stage cascaded architecture. Its key innovation lies in synergistically combining adaptive statistical denoising with geometry-semantic joint optimization—specifically, enabling dynamic adjustment of the Z-score threshold in coordination with hierarchical plane classification for the first time. Evaluated on a real-world architectural point cloud dataset, the framework achieves a plane detection recall of 92.1% and a semantic segmentation mean Intersection-over-Union (mIoU) of 89.3%, significantly improving recognition accuracy for architectural components (e.g., walls, windows, doors) and advancing automation in 3D reconstruction.

Estimating surface normals from noisy point clouds remains a longstanding challenge, as existing methods rely heavily on clean inputs and strong supervision, failing to simultaneously preserve geometric details and ensure robustness. This paper proposes a local gradient-aware implicit surface filtering framework: it guides point cloud projection onto the underlying surface via an implicit function and introduces a joint normal-distance operator to enforce geometric constraints. To mitigate gradient degradation, we design a local gradient consistency loss; to suppress over-smoothing, we develop a multi-scale directional aggregation mechanism. Crucially, our method performs end-to-end learning without explicit normal supervision, unifying denoising, normal estimation, and surface reconstruction. Extensive experiments demonstrate state-of-the-art performance across multiple benchmarks, significantly improving fine-detail recovery under complex noise patterns. The code and pre-trained models are publicly available.

To address the limitations of existing point cloud surface reconstruction methods—namely, reliance on normal vectors, fixed control point counts, and insufficient robustness to noise—this paper proposes a dictionary-guided graph convolutional B-spline surface reconstruction framework. Our method eliminates the need for input point normals and jointly learns both the number and 3D positions of B-spline control points in an end-to-end manner, leveraging a learnable control point dictionary integrated with a graph neural network. This enables adaptive geometric complexity modeling. The resulting architecture inherently supports topological flexibility while preserving geometric continuity. Quantitative evaluations on multiple benchmark datasets demonstrate significant improvements over state-of-the-art approaches in standard metrics—including Chamfer distance and F-Score—as well as superior qualitative visual fidelity. Notably, our method exhibits enhanced robustness under high-noise conditions, achieving more accurate and stable surface reconstructions.

This work addresses the failure of 5-point and 7-point RANSAC-based relative pose estimation in multiview geometry—even under outlier-free conditions and sufficient data—identifying inherent numerical instability in minimal problems as the root cause. To tackle this, we propose a unified conditioning analysis framework that integrates computational algebra (Gröbner bases, elimination theory) with Riemannian geometry (condition number definition on manifolds), enabling the first computable modeling and scene-level predictability of condition numbers for the 5- and 7-point problems. Theoretically, we reveal a previously unrecognized mechanism: RANSAC implicitly favors well-conditioned image configurations due to its sampling and scoring behavior. Extensive experiments confirm that this implicit bias is essential for RANSAC’s success in outlier-free settings and demonstrate substantial improvements in pose estimation robustness and reliability across diverse geometric scenarios.

This work addresses the challenge of degraded 3D surface reconstruction accuracy caused by missing geometric information in LiDAR point clouds due to limited scanning range and occlusions. To tackle this issue, the authors propose a reconstruction method based on plane classification and priority-driven growth. The approach categorizes scene planes into three visibility classes—highly visible, partially visible, and invisible—and employs a hierarchical spatial partitioning scheme. Coupled with a min-cut optimization strategy, it generates compact, watertight polygonal models that effectively recover missing geometric details. Evaluated on public datasets, the method significantly outperforms current state-of-the-art techniques, achieving higher reconstruction fidelity while preserving model compactness.

This work proposes a parameter-free local topographic descriptor for the comparison and rigid alignment of three-dimensional structured point patterns. The method decomposes each point pattern into multiple arms and introduces a normalized finite difference operator along each arm to capture the local variation of height components relative to the underlying planar geometry, thereby integrating fine-grained geometric details with global structural information. By combining Wasserstein distance with Procrustes analysis, the approach enables efficient distributional comparison and precise alignment of point clouds. The proposed descriptor preserves salient local topographic features while significantly enhancing the robustness and accuracy of point pattern matching.

This work proposes a novel method for jointly estimating an implicit surface and its noise distribution from noisy measurements. The surface is modeled as the zero level set of a function in a reproducing kernel Hilbert space, while a parametric noise model is introduced to capture measurement uncertainty. The resulting optimization framework requires no hyperparameter tuning or preprocessing and scales linearly with data size, offering theoretical convergence guarantees. Implicit surface representations are constructed using polynomial and trigonometric basis functions, enabling applicability to high-dimensional data. Extensive experiments on 2D and 3D benchmark datasets demonstrate that the proposed approach consistently outperforms state-of-the-art methods such as Poisson reconstruction and Encoder-X in both reconstruction accuracy and computational efficiency.

This work proposes a robust watertight surface reconstruction method for unoriented, non-uniformly sampled point clouds corrupted by noise and outliers. The approach uniquely integrates the joint optimization of point normals, area weights, and confidence coefficients within a unified framework, constructing an implicit representation based on the generalized winding number field. Smoothness is enforced through Dirichlet energy minimization, providing effective regularization without requiring any preprocessing to handle complex point cloud defects. Experimental results demonstrate that the method consistently produces high-quality watertight surfaces from challenging inputs, including outputs from 3D Gaussian Splatting and corrupted geometry benchmarks, outperforming both conventional multi-stage pipelines and existing joint reconstruction techniques.

This work addresses the challenge of severe depth sensor distortions caused by glass surfaces in indoor robot navigation, which compromise environment reconstruction and path planning. We propose a training-free fusion framework that, for the first time, leverages foundation depth models—such as Depth Anything V3—as structural priors, integrated with a local RANSAC alignment strategy to effectively reject erroneous depth measurements in glass regions and recover metrically accurate geometry. To support this research, we introduce GlassRecon, the first RGB-D dataset featuring geometrically derived ground-truth annotations for glass surfaces. Experimental results demonstrate that our method significantly outperforms existing approaches in depth distortion scenarios, achieving accurate glass geometry reconstruction while preserving metric consistency. Both the code and the GlassRecon dataset are publicly released.