Hierarchical active learning for annotation-efficient 3D object detection in construction-scene LiDAR sensing

3D object detection in construction-scene LiDAR sensing remains challenging because sparse returns, irregular layouts, heavy occlusion, and diverse machinery make object representation and localization difficult. Dense manual annotation in such scenes is also inefficient and costly, creating a strong need for annotation-efficient learning. Existing active-learning methods often struggle in this setting because they tend to favor uncertain but redundant samples while overlooking representative scene coverage. To address this problem, we propose Hierarchical Clustering-enhanced Active Learning (HCAL), an active-learning framework for LiDAR-based 3D object detection in construction and roadwork environments. HCAL improves sample selection under limited labeling budgets by identifying informative candidate keyframes, suppressing redundant uncertain samples, and preserving representative coverage through entropy filtering, gradient-space clustering, and density-consistency selection. To support field-relevant evaluation, we further establish ConVeh3D as a construction-scene LiDAR benchmark for annotation-efficient 3D detection and unknown-category-aware evaluation in a controlled category-withheld setting. Experiments on ConVeh3D, with KITTI as secondary supporting evidence, show that HCAL achieves the best average detection accuracy across most tested labeling budgets. The results indicate that HCAL provides an effective framework for improving sample efficiency in construction-scene LiDAR-based 3D object detection under limited labeling budgets. The dataset and code will be made available at https://github.com/zxk688/ConVeh3D .

Authors

Institutions

Publication Details

Journal
International Journal of Applied Earth Observation and Geoinformation
Published
2026-11-16
DOI
https://doi.org/10.1016/j.jag.2026.105678
Primary Topic
Advanced Neural Network Applications
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Hierarchical active learning for annotation-efficient 3D object detection in construction-scene LiDAR sensing

Jialu Sui, Xiaokang Zhang, You Li, Yasong Shi et al.
International Journal of Applied Earth Observation and Geoinformation
Advanced Neural Network Applications
article

Hierarchical active learning for annotation-efficient 3D object detection in construction-scene LiDAR sensing

Jialu Sui, Xiaokang Zhang, You Li, Yasong Shi, Renzhong Guo, Shengjun Tang, Ruisheng Wang
article en

Abstract

3D object detection in construction-scene LiDAR sensing remains challenging because sparse returns, irregular layouts, heavy occlusion, and diverse machinery make object representation and localization difficult. Dense manual annotation in such scenes is also inefficient and costly, creating a strong need for annotation-efficient learning. Existing active-learning methods often struggle in this setting because they tend to favor uncertain but redundant samples while overlooking representative scene coverage. To address this problem, we propose Hierarchical Clustering-enhanced Active Learning (HCAL), an active-learning framework for LiDAR-based 3D object detection in construction and roadwork environments. HCAL improves sample selection under limited labeling budgets by identifying informative candidate keyframes, suppressing redundant uncertain samples, and preserving representative coverage through entropy filtering, gradient-space clustering, and density-consistency selection. To support field-relevant evaluation, we further establish ConVeh3D as a construction-scene LiDAR benchmark for annotation-efficient 3D detection and unknown-category-aware evaluation in a controlled category-withheld setting. Experiments on ConVeh3D, with KITTI as secondary supporting evidence, show that HCAL achieves the best average detection accuracy across most tested labeling budgets. The results indicate that HCAL provides an effective framework for improving sample efficiency in construction-scene LiDAR-based 3D object detection under limited labeling budgets. The dataset and code will be made available at https://github.com/zxk688/ConVeh3D .

International Journal of Applied Earth Observation and GeoinformationVol. 155
Shenzhen University (CN), Wuhan University (CN), Wuhan University of Science and Technology (CN), Guangdong Laboratory of Artificial Intelligence and Digital Economy (Shenzhen) (CN), University of Hong Kong (HK)
Advanced Neural Network Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon