Robust Reconstruction of Power Lines from Incomplete Mobile and Aerial LiDAR Data
Authors: Mohamed Gaha, Damos Ayobo Abongo, Philippe Massicotte, Eya Ben Kacem, Wael Jaafar, Rami Langar, Christian Buteau
Organizations: Hydro-Quebec, QC J3X 1S1, Canada · Software and IT Engineering Department, École de Technologie Supérieure, Montreal, QC H3C 1K3, Canada · LIGM-CNRS UMR 8049, University Gustave Eiffel, F-77420 Marne-la-Vallée, France
Accurate vegetation-encroachment assessment and risk mitigation require electric utilities to accurately locate and characterize distribution-network assets, including poles and power lines, and distinguish among medium-voltage, low-voltage, neutral, and telecommunication lines. Mobile and aerial Light Detection and Ranging (LiDAR) point clouds provide detailed 3D representations of these assets and their surroundings. However, power-line detection and reconstruction remain challenging because conductor returns are often sparse, incomplete, noisy, or occluded by vegetation. These challenges are especially significant in distribution networks, where lower pole heights and complex line configurations place conductors close to trees. In addition, mobile and aerial LiDAR datasets differ substantially in point density, viewing geometry, spatial coverage, and occlusion patterns. We propose a robust and automated reconstruction method combining scaled density-based clustering, robust polynomial fitting, and iterative fragment merging to recover individual conductors under severe data incompleteness and occlusion. Following semantic classification, the same reconstruction pipeline processes both mobile and aerial LiDAR data without modality-specific code or parameter tuning. The method is evaluated through degradation experiments, benchmarking, parameter sensitivity analysis, and large-scale validation. On the aerial dataset, it achieves 96.87% precision, 96.66% recall, a 96.76% F1-score, and a geometric RMSE of 0.127 m. On the mobile dataset, where conductor-level reference annotations were unavailable, it achieves a point-to-model RMSE of 0.045 m and a reconstruction coverage of 95.27%. Source code, documentation, and evaluation data are publicly available at https://github.com/varen-project.
Figures & tables
Reference
Objective
Main technique
Context or model
Data
Reported performance
Main limitation
Jwa et al. Jwa and others (2009)
3D power-line reconstruction
Voxel-based piecewise detection and catenary fitting
Local voxel grouping and iterative catenary refinement
ALS
Accuracy: 93.8%
Sensitive to vegetation-induced occlusion and potentially computationally expensive
Liang and Zhang Liang and others (2011)
Power-line extraction and modeling
Geometric filtering, Hough-like detection, and least-squares fitting
Three-dimensional parabolic conductor model
ALS
Accuracy: 96.5%
Performance deteriorated when conductor observations contained large gaps
Guo et al. Guo and others (2016)
Robust conductor reconstruction
Cuboid segmentation, quadratic fitting, and iterative RANSAC
Span constrained by two pylons
ALS
Accuracy: 94.9%
Assumed a sufficiently consistent conductor configuration within each span
Gaha et al. Gaha and others (2021)
Pole and power-line detection
RANSAC-based segment extraction and RGB filtering
Plane defined by consecutive poles and three-point modeling
MLS
Recall: 98%; accuracy: 84%
Limited to single-phase configurations and sensitive to candidate-point errors
Shi et al. Shi and others (2020)
Urban power-line extraction and hazard analysis
Geometric feature extraction and clustering
Urban roadside distribution-network context
MLS
Accuracy: 91%
Designed for mobile LiDAR and not evaluated on aerial data
Massicotte et al. Massicotte and others (2024)
Utility-asset inventory
Combined image and LiDAR processing
Contextual detection of utility assets
MLS
Accuracy: 85–95%
Limited implementation details restricted reproducibility and analysis
Table 1: Comparison of LiDAR-based power-line detection and reconstruction methods.
Figure 1: Example of a mobile LiDAR span classified using U-Next. Vegetation is shown in green, power lines in light blue, poles in dark blue, vehicles in orange, public-lighting assets in light green, and other objects in red.
Class
Train (%)
Val. (%)
Test (%)
Count (M)
Vegetation
89.79
72.54
89.98
134.6
Wire
1.31
3.24
1.54
2.6
Pole
0.98
2.25
1.16
1.9
Transformer
0.04
0.11
0.03
0.07
Building
5.71
17.29
6.26
11.9
Vehicle
1.38
1.69
0.83
2.0
Table 2: Class distribution in the HQ-2025 mobile LiDAR dataset. Percentages indicate the proportion of points within each split.
Figure 2: Conductor fragments obtained using span-aligned prescaling, HDBSCAN clustering, and geometric filtering. Each color represents a different cluster.
Figure 3: Reconstruction F 1 -score for three conductor architectures under controlled degradation: uniform point thinning, multiple distributed gaps, and a single contiguous gap.
Figure 4: Reconstruction of a three-phase span using 5% of the original conductor observations (red: retained LiDAR observations, white: reconstructed conductor).
Figure 5: Example of a single contiguous occlusion covering 75% of the conductor span (red: retained LiDAR observations, white: reconstructed conductor).
Reference
Predicted
TP
FP
FN
Precision (%)
Recall (%)
F 1 (%)
RMSE (m)
23,372
23,323
22,592
731
780
96.87
96.66
96.76
0.129
Table 3: Large-scale performance on the aerial LiDAR dataset.
Metric
Mean
Std
Median
P95
Max
F 1 per span (%)
96.31
15.82
100.00
100.00
100.00
Processing time per span (s)
0.153
0.342
0.100
0.350
0.47
Point-to-reference distance (m)
0.056
0.091
0.040
0.112
3.00
RMSE per conductor (m)
0.095
0.072
0.073
0.234
1.04
Table 4: Distribution of the per-span F 1 -score, processing time and geometric error on the aerial LiDAR dataset.
Figure 6: Comparison of the proposed method, EPL, and Cao-FEC (mod.) under a single contiguous gap for mobile LiDAR.
Figure 7: Comparison of the proposed method, EPL, and Cao-FEC (mod.) under uniform point thinning for mobile LiDAR.
Figure 8: Comparison of the proposed method, EPL, and Cao-FEC (mod.) under multiple distributed gaps for mobile LiDAR.
Parameter
Description
Selected value
Type
ϵHDBSCAN
HDBSCAN cluster-selection epsilon in the scaled coordinate space
0.3
Fixed
kmin
HDBSCAN minimum number of samples
3
Fixed
mcs
HDBSCAN minimum cluster size depending on points density
[3,5]
Adaptive
wn
Scale weight applied to the cross-span coordinate
2
Fixed
wz
Scale weight applied to the vertical coordinate
1
Fixed
ws,min
Lower bound on the adaptive along-span scale weight
0.05
Fixed
Table 5: Selected parameters of the proposed reconstruction pipeline.
ϵHDBSCAN
TP
FP
FN
Prec.
Rec.
F 1
0.1
402
4
18
99.0
95.7
97.3
0.2
400
7
20
98.3
95.2
96.7
0.3
402
6
18
98.5
95.7
97.1
0.4
402
6
18
98.5
95.7
97.1
0.5
400
9
20
97.8
95.2
96.5
0.7
399
7
21
98.3
95.0
96.6
Table 6: Sensitivity to HDBSCAN cluster-selection epsilon for mobile LiDAR.
τRMSE
TP
FP
FN
Prec.
Rec.
F 1
0.05
401
24
19
94.4
95.5
94.9
0.10
401
9
19
97.8
95.5
96.6
0.15
402
8
18
98.0
95.7
96.9
0.20
403
5
17
98.8
96.0
97.3
0.25
401
6
19
98.5
95.5
97.0
0.30
402
6
18
98.5
95.7
97.1
Table 7: Sensitivity to the Huber RMSE merging threshold for mobile LiDAR.
Table 8: Influence of the curve model and fitting estimator for mobile LiDAR.
Standard RMSE
Huber RMSE
τRMSE
TP
FP
FN
F 1
TP
FP
FN
F 1
0.10
401
9
19
96.6
401
9
19
96.6
0.15
401
8
19
96.7
402
8
18
96.9
0.20
403
5
17
97.3
403
5
17
97.3
0.25
403
5
17
97.3
401
6
19
97.0
0.30
403
5
17
97.3
402
6
18
97.1
Table 9: Standard versus Huber RMSE merging criterion for mobile LiDAR. The adopted configuration is the Huber RMSE with τRMSE=0.30 m.
Dataset
Configuration
TP
FP
FN
Prec.
Rec.
F 1
Mobile
Prescaling
402
6
18
98.5
95.7
97.1
No prescaling
400
5
20
98.8
95.2
97.0
Table 10: Detection performance with and without span-aligned prescaling for mobile and aerial LiDAR. Precision, recall, and F 1 -score are reported as percentages.
Param.
Selec. Val.
Test. range
Min. F 1
Max. F 1
τRMSE
0.30
[0.05,1]
19.9 %
97.3 %
δH
0.10
[0.02,0.5]
19.9 %
97.3 %
τspan
0.55
[0.2,1]
90.9 %
97.1 %
ws,max
1
[0.1,2]
97.1 %
97.1 %
ws,min
0.05
[0.01,0.5]
97.1 %
97.5 %
ϵHDBSCAN
0.3
[0.1,2]
96.5 %
97.5 %
Table 11: Summary of the one-at-a-time parameter sensitivity analysis for mobile LiDAR.
This paper presents an end-to-end framework for reconstructing overhead power utility network topology and extracting span-level physical metadata from large-scale aerial LiDAR. The pipeline begins with semantic segmentation of the input point cloud using an improved KPConv-based model, in which data sampling and loss functions are adapted to emphasize pole and conductor (wire) classes. Network topology inference then proceeds in two stages: (i) pole instances are obtained by clustering pole-class points and validating candidates using geometric criteria, including height and verticality estimated via PCA, and (ii) candidate pole pairs are evaluated using a heuristic method and a lightweight ResNet-based classifier on 2D top-view projections of pole and wire point distributions to determine whether a physical conductor span exists. By explicitly classifying candidate spans, the approach mitigates common failure modes of heuristic connectivity rules in dense or cluttered scenes and under partial wire observation. For each validated wire, attributes regarding utility infrastructure geometry are computed, including endpoint conductor heights, ground elevation, sag-related lowest-point features, conductor arrangement, and wire width. Evaluation on multiple real-world aerial LiDAR datasets demonstrates decimeter-level endpoint height accuracy and approximately 9% relative improvement in recall for topology reconstruction compared to heuristic nearest-neighbor baselines, with larger gains in complex layouts.
This article deals with the problem of detecting the position of power lines in airborne laser scanning point clouds. We address both the problem of segmenting power lines from background like ground and vegetation and of identifying individual lines. For segmentation we compare a rule-based approach with one based on transfer learning (Point-SAM). For individual line detection, we derive the mathematical formulas for RANSAC applied to 3D catenaries and compare it with a non-parametric method (TriplClust). All methods are evaluated on a newly created dataset "PowLiDFR", which we make publicly available. The samples have been taken from the French project LiDAR HD and cover a variety of terrain layouts and point cloud resolutions. On this dataset, an unsupervised transfer learning outperformed a mere rule-based segmentation, and for individual power line detection, RANSAC was better in following the shape of catenaries than the non-parametric algorithm.
Christoph Dalitz, Oskar Waedt, Steffen Goebbels +2
Institute for Pattern Recognition, Niederrhein University of Applied Sciences, Reinarzstr. 49, Krefeld, Germany · Forestry France, 16 ter bd de la Taillerie, 63130 Royat, France
Ground robots operating in temperate forests suffer from systematic LiDAR occlusions caused by tree trunks, creating persistent blind regions that degrade traversability estimation. In this paper, we propose FOSCo-Net, a self-supervised voxel completion framework that learns to reconstruct occluded geometry from a single scan using temporally accumulated observations as automatic supervision, and we evaluate cylindrical and Cartesian ego-centric grids with multiple decoder designs. FOSCo-Net consistently improves reconstruction quality and traversal safety, with a gated skip decoder achieving the most reliable performance and significantly reducing false obstacle predictions compared to baselines.
Iman Esfandiyar, Dominik Belter
Poznań University of Technology, Institute of Robotics and Machine Intelligence, Poznań, Poland · Łukasiewicz Research Network – Poznań Institute of Technology, Center of Digital Transformation, Department of Artificial Intelligence, Poznań, Poland