• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Wednesday, November 18, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Technology

Information-Aware Reduction Preserves Minority Classes in Construction LiDAR Scans

by
November 18, 2026
in Technology
Reading Time: 3 mins read
0
Information-Aware Reduction Preserves Minority Classes in Construction LiDAR Scans

Information-Aware Reduction Preserves Minority Classes in Construction LiDAR Scans

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

In a systematic evaluation of 51 terrestrial laser-scan scenes from two Rohbau3D construction sites, researchers demonstrated that balanced information-aware selection preserves critical minority elements while maintaining necessary data decimation rates. The investigation highlights how conventional global distortion metrics often obscure severe data losses among sparsely represented structural categories during routine site data processing.

The study established a controlled testing framework where every analytical outcome was conditional on a class-aware reference sample of 150,000 points per scene, totaling 7.65 million points, rather than native full-resolution scans. The investigative team evaluated multiple reduction strategies against a naïve NPZ and Deflate storage baseline, which reduced the sampled payload size by 10.13 percent. In addition to storage compression, the benchmark examined voxel grid filtering across resolutions of 0.05, 0.10, and 0.20 meters, alongside five archived fixed-rate selection policies tested at retention levels of 25 percent, 10 percent, and 5 percent.

Testing across the voxel grid configurations revealed monotonic changes across all measured outcomes as cell dimensions expanded. Applying a 0.05-meter voxel grid successfully eliminated 69.83 percent of the sampled points while generating a mean nearest-neighbour error of 1.470 centimeters and maintaining a 99.32 percent nearest-neighbour class agreement. However, coarser voxel filters exposed critical limitations regarding minority elements. When the voxel grid was widened to 0.20 meters, 83.77 percent of rare class–scene records retained less than 10 percent of their sampled points, although no evaluated class disappeared entirely from the records.

Geometric registration stability showed distinct trade-offs under voxel filtering during controlled Iterative Closest Point evaluations. Implementing a 0.05-meter voxelisation did not cause a statistically significant change in the archived control root-mean-square error measure when compared directly to the sampled-reference condition. Despite this stability in overall error, the proportion of surface points achieving fitness within 10 centimeters declined sharply by 13.62 percentage points, a shift verified as statistically significant at p less than .001.

In comparative testing of fixed-rate downsampling algorithms, balanced information-aware selection delivered consistent advantages over standard class-aware selection routines. While both strategies operated under identical class quotas and therefore maintained identical rare-class retention levels, the balanced approach significantly bolstered small-instance presence across all evaluated thresholds. Specifically, balanced information-aware selection improved small-instance presence over class-aware selection by 2.10 percentage points at 25 percent retention, 7.21 percentage points at 10 percent retention, and 15.47 percentage points under an aggressive 5 percent retention setting.

To ensure robust conclusions across diverse scanning environments, the research team replaced conventional point-pooled statistical inference with scene-level bootstrap confidence intervals and repeated-measures tests. The resulting metrics track the actual information retained within construction point clouds following geometric reduction, providing quantitative measurements that operate on top of downstream semantic segmentation or automated object-detection performance.

By detailing the specific trade-offs between scene-level fidelity, minority-class preservation, and geometric registration precision, the study delivers a transparent, construction-specific benchmark for engineering workflows.

Subject of Research: Information-aware reduction of construction LiDAR point clouds

Article Title: Information-Aware Reduction of Construction LiDAR Point Clouds: Scene-Level Fidelity, Minority-Class and Registration Trade-offs

Article References: Banihashemi, S., Kalantari, M., & LeNguyen, K. (2026). Information-Aware Reduction of Construction LiDAR Point Clouds: Scene-Level Fidelity, Minority-Class and Registration Trade-offs. Journal of Intelligent & Robotic Systems. https://doi.org/10.1007/s10846-026-02490-x

Image Credits: AI Generated

DOI: 10.1007/s10846-026-02490-x

Keywords: LiDAR point clouds, terrestrial laser scanning, point cloud reduction, Rohbau3D, voxel grid filtering, iterative closest point, rare class retention, minority-class preservation, small-instance presence, construction point clouds, geometric registration, nearest-neighbour error, scene-level fidelity, fixed-rate selection, information-aware reduction, class agreement, control-RMSE, NPZ Deflate storage

News Source: Denise Maddox. (November 18, 2026). Information-Aware Reduction Preserves Minority Classes in Construction LiDAR Scans. Scienmag.

Tags: class agreementconstruction point cloudscontrol-RMSEfixed-rate selectiongeometric registrationinformation-aware reductioniterative closest pointLiDAR point cloudsminority-class preservationnearest-neighbour errorNPZ Deflate storagepoint cloud reductionrare class retentionRohbau3Dscene-level fidelitysmall-instance presenceterrestrial laser scanningvoxel grid filtering
Share12Tweet7Share2ShareShareShare1

Related Posts

Coal Gangue Aggregates Can Replace Natural Stone in Concrete Applications

Coal Gangue Aggregates Can Replace Natural Stone in Concrete Applications

November 18, 2026
Cluster-Based Training Selection Outperforms Active Learning for Material Property Prediction

Cluster-Based Training Selection Outperforms Active Learning for Material Property Prediction

November 18, 2026

LLM-Assisted Network Improves Federated Graph Learning and Privacy in IIoT

November 18, 2026

Decoupled Multiplex Causal Ranking Improves Microservice Root Cause Localization in Benchmarks

November 18, 2026

POPULAR NEWS

  • Coal Gangue Aggregates Can Replace Natural Stone in Concrete Applications

    Coal Gangue Aggregates Can Replace Natural Stone in Concrete Applications

    29 shares
    Share 12 Tweet 7
  • Cluster-Based Training Selection Outperforms Active Learning for Material Property Prediction

    29 shares
    Share 12 Tweet 7
  • Cesarean Section Rates Reached Forty Percent Among Surveyed Mothers in Njombe Region

    29 shares
    Share 12 Tweet 7
  • Testosterone Injections Exacerbate Pre-Neoplastic Prostate Lesions in NKX3.1-Deficient Mice

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Coal Gangue Aggregates Can Replace Natural Stone in Concrete Applications

Cluster-Based Training Selection Outperforms Active Learning for Material Property Prediction

Cesarean Section Rates Reached Forty Percent Among Surveyed Mothers in Njombe Region

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 88 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.