Reconstruction of interior walls from point cloud data with min-hashed J-linkage

Oct 24, 2026

Upload by Putri

Automatic reconstruction of interior walls is central to scan2BIM systems and requires both robustness and efficiency when processing large point clouds. This work introduces an improved J-Linkage variant that integrates min-Hash to accelerate computation without losing accuracy. A complete framework extracts floor plans from large-scale point clouds and is validated on real-world, publicly available datasets. The approach targets reliable separation of planar structures despite noise and outliers.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Introduction

J-Linkage and min-Hash

J-Linkage algorithm

Reconstruction of interior walls from point cloud data with min-hashed J-linkage Luca Magri, Andrea Fusiello DPIA-University of Udine Via delle Scienze, 208, -33100 Udine, IT [surname.name.l@gmail.com](surname.name.l@gmail.com) , [name.surname@uniud.it](name.surname@uniud.it) Abstract The automatic reconstruction of the walls of an interior environment is a fundamental task in any “scan2BIM” application. In this work, we address this problem resorting to an original and improved version of J-Linkage that leverages on the min-Hash technique to boost the efﬁciency without sacriﬁcing the accuracy. A framework to automatically and robustly extract ﬂoor plans from large-scale point clouds is described and validated on real-word publicly available data. 1. Introduction Understanding the structure of indoor environments is an important task that, in recent years, has attracted the interest of researchers yielding to several applications that span many ﬁelds such augmented and virtual reality, navigation, design, maintenance, monitoring of buildings and energy simulation analysis. Speciﬁcally, the increasing availability of high quality and massive 3D data, the diffusion ofBIM and the sustained need of accurate as-built architectural models has driven the attention towards automatic methods capable of dealing with large-scale point clouds and, at the same time, able to retrieve detailed geometry, avoiding time-consuming and costly human intervention as much as possible. The demand of higher-level, automated and accessible rendition of indoor models, can be regarded as an occurrence of the broader aim of bridging the semantic gap that separates automatic perception from human comprehension which ultimately inspires many efforts in Computer Vision. A ﬁrst preliminary step along this direction is taken by geometric multi-structure recovery (also known as multi-model ﬁtting), which aims at aggregating input data, usually corrupted by noise and outliers, into multiple instances of geometric parametric models. In the context of indoor modeling, this translates mainly in the robust extraction of multiple geometric primitives consisting of the primary facility surfaces – such as ﬂoors, walls, and ceilings – from the point cloud. Usually planar structures are detected exploiting 3D plane ﬁtting techniques e.g. [20] . Alternatively, the description of the environment is conveniently addressed from a 2D perspective and the architectural components are described as a mixture of multiple line segments, as in a blueprint e.g. [18] . Eventually, under the assumption of vertical surfaces, a 3D model can be easily extruded from the blueprint. The typical challenges of multi-model ﬁtting are lurking here. The structure estimation process must tolerate pseudo-outliers and rogue measurements, due to the presence of furniture causing clutter and complex building layouts which determine occlusions. Moreover, the ﬁt of geometric models and the assignment of points to the estimated structures are intrinsically entangled tasks and give rise toa sort of chicken-and-egg-dilemma. In addition, the number of sought structures is unknown in advance and must be estimated together with the parametric models, entailing implicitly or explicitly some model selection criterion. Multiple structure recovery is a common essential step involved in the majority of the works about indoor modeling, including the most elaborate methods, which make use of it as a preparatory phase to further semantic or topological analysis. Some long-established algorithms are customarily adopted: [18, 1, 25] rely on Hough transform [9] to detect collinear conﬁgurations of 2D wall samples. Unfortunately this technique is not very robust to clutter and outliers, especially as the number of structures grows and the distribution of inliers per structure is uneven. Along the same line, [14] resorts to mean-shift [7], whose parametrization can be critical and which is not

FAQs

What problem does the paper address?

It addresses automatic reconstruction of interior walls from point cloud data for scan2BIM applications, aiming to extract floor plans robustly from large-scale point clouds.

How does the proposed method improve J-Linkage?

It tailors J-Linkage by leveraging the min-Hash technique to improve efficiency while maintaining accuracy.

How is the method validated?

The framework is described and validated on real-world publicly available data.

Oct 24, 2026

Upload by Putri