DOI: 10.3303/CET25121016 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Paper Received: 23 August 2025; Revised: 9 November 2025; Accepted: 05 December 2025 
Please cite this article as: Oláh S., Barsi Á., Kozma K., 2025, Potential of Point Cloud Upsampling for Environmental Protection: Enhancing 
Airborne LiDAR Data for Sustainable Resource Management, Chemical Engineering Transactions, 121, 91-96  DOI:10.3303/CET25121016 
  

 CHEMICAL ENGINEERING TRANSACTIONS  
 

VOL. 121, 2025 

A publication of 

 

The Italian Association 
of Chemical Engineering 
Online at www.cetjournal.it 

Guest Editors: Petar S. Varbanov 

Copyright © 2025, AIDIC Servizi S.r.l. 

ISBN 979-12-81206-22-9; ISSN 2283-9216 

Potential of Point Cloud Upsampling for Environmental 

Protection: Enhancing Airborne LiDAR Data for Sustainable 

Resource Management 

Szeverin Oláha, Árpád Barsib, Katalin Kozma*,c 

aDoctoral School of Multidisciplinary Engineering Sciences, Széchenyi István University, Egyetem tér 1, 9026 Győr, 

 Hungary 
bDepartment of Photogrammetry and Geoinformatics, Faculty of Civil Engineering, Budapest University of Technology and  

 Economics, Műegyetem rkp. 3, 1111 Budapest, Hungary 
cDepartment of Applied Sustainability, Albert Kázmér Faculty of Agricultural and Food Sciences, Széchenyi István 

 University, Egyetem tér 1, 9026 Győr, Hungary 

 kozma.katalin@sze.hu 

High-resolution, dynamic geospatial data support sustainable infrastructure, optimize urban services, and 

improve the quality of life of residents. Airborne Laser Scanning is an increasingly popular remote sensing 

technology that can be used to collect very large datasets of 3D point clouds over extensive areas, including 

forests, river basins, coastal wetlands, and mountainous regions. These datasets facilitate the analysis of 

vegetation structure, biomass estimation, hydrological modeling, and land cover detection or change monitoring. 

However, Airborne Laser Scanning-derived point clouds are typically limited to low density and spatial 

resolution, which preclude meaningful analysis for fine-scale ecological and environmental modeling.  

Point cloud upsampling is a permissible way to augment the spatial robustness of Airborne Laser Scanning 

point cloud data, and does so without adding a logistical burden of the data acquisition in the field, or the need 

to resurvey at high costs and time. Upsampling is synthetic in nature, achieving increased data point count, but 

maintaining dimensional integrity for continuity of surfaces and geometric fidelity, which is essential in 

methodologies that intervene for derived products such as digital terrain models, canopy height models, and 

vegetation metrics. This manuscript examines using point cloud upsampling as part of environmental monitoring. 

It reviews the upsampling algorithms that have been developed to date, synthesizes existing methods, and 

considers their relevance to the state of practice in forestry, watershed management, and conservation planning. 

The work considers and focuses on methodological bases for robustness and dimensionality, and although 

considerably nuanced, the methodological efficacy is subtended and suggests how enhanced points improve 

outcomes for ecological models and the information provided supports resource management decisions related 

to resource and sustainability decisions.. Ultimately and conclusively, the work establishes the understanding 

of point cloud enhancement for its visualization, but also its potential as an emergent action that contributes to 

construction and promotes a sustainability intention in environmental science and policy. 

This article appears as a mini-review. This writing aims to synthesise existing knowledge and conceptual 

strategies, rather than a novel outcome of an experiment. It is written to give an overview of existing methods 

and structured conceptual frameworks for employing point cloud upsampling techniques in the environmental 

monitoring and sustainability context. The review reiterates the conceptual soundness of point cloud upsampling 

in the workflow of environmental monitoring. The proposed framework reinforces the benefits of greater data 

richness and decision-making based on sustainability, without assuming new costs for data generation. 

1. Introduction 

The monitoring, management, and conservation of natural environments are becoming an increasingly serious 

issue due to global climate change, rapid loss of biodiversity, and added anthropogenic pressures to ecosystem 

processes. In light of these pressures, the world will need high-quality spatial data that is timely, accurate, and 

91



high-resolution to support the development of reasonable decision-making, effective policy design, and 

sustainable resource management decisions. According to Skidmore et al. (2015), it is essential to have 

biodiversity metrics to help consistently track biodiversity from space, and Venter et al. (2016) demonstrate the 

ways in which long-term changes in the global human footprint can impact the priority for biodiversity and 

resource management. Fine-scale observations are essential for monitoring ecosystem processes, planning 

conservation strategies or ecosystem management planning. In contrast to earlier reviews, which focused solely 

on geometric or algorithmic efficiency, this article frames point cloud upsampling within environmental 

sustainability and monitoring. Airborne Laser Scanning (ALS) is an example of remote sensing technologies 

and is a valuable tool for collecting three-dimensional data over a large area in environments that are often too 

difficult to access or understudied, such as forests, mountains, and coastal environments. For example, Hollaus 

et al. (2007) showed the potential of ALS for detailed structural characterizations of forests in rugged 

mountainous environments, and White et al. (2016) showed the potential for ALS to inform extensive forest 

inventories and resource management.  

Most ALS systems used today use near-infrared wavelengths. They are mounted on airplane platforms 

(helicopters, drones, balloons, etc.) flying at altitudes of several hundred m, up to above 1,000 m, depending on 

mission design parameters and regulations. Contemporary ALS sensors can produce point cloud (PC) densities 

of 5–30 points per square metres (pts m²), with typical mean densities between 10 and 15 pts m² for forest 

inventory applications. Current vertical location accuracies are usually about ±5–10 cm, while horizontal 

(planimetric) accuracies are usually within 0.2 m. With regular acquisition and processing conditions, the root 

mean square error (RMSE) of interpolated surface models of the ground is generally between 0.10 m and 

0.30 m, with variability based on terrain morphology, vegetation structure, and the interpolation used. 

In related fields, more robust classical registration schemes such as the variant of the Iterative Closest Point 

(ICP) have been suggested, which can improve the robustness and the geometric fidelity of PC data types (Sha 

et al., 2016).  If this PC data is sparse, digital terrain models (DTMs), Canopy height models (CHMs), vegetation 

metrics, etc., are limited in their accuracy and ultimately limit the rigor of relevant models/assessments that are 

based on such data. Specifically, deep learning (DL) based methods have the potential to synthetically augment 

PCs by increasing the point density of the PC area but maintaining the basic geometric forms and surface 

continuity. Traditional interpolation techniques, especially large area ones, tend to create artifacts or oversmooth 

features, whereas a DL based upsampling method can capture complex patterns and fine-scale variability 

apparent in natural settings. DL proposed methods for upsampling can increase density in a paranormal way 

while still retaining surface geometric fidelity and continuity. Accordingly, environmental modeling and feature 

extraction can be enhanced in a manner that involves no additional fieldwork or expensive resurveys (Zhang 

and Filin, 2022). This is particularly advantageous in the case of operational approaches to forest inventory, 

watershed hydrology modelling, monitoring coastal erosion, or conserving habitats, where the quality of spatial 

resolution and richness of data absolutely affect the effectiveness of management actions. The primary aim of 

the research was to investigate point cloud upsampling (PU) and its function in improving sustainable 

management through environmental monitoring. In this paper, we provide a comprehensive overview of existing 

PU algorithms, which mainly include classic and DL approaches. This paper is presented as a mini-review and 

conceptual study. Instead of offering new experimental results, the structure summarises the research 

landscape PU for ALS data and offers a conceptual structure to help guide future use in environmental 

monitoring and sustainability analysis. 

2. Background and related work 

Managing the environment sustainably and responsibly requires access to accurate and fine-scale spatial data 

that can adequately reflect the complexity of ecological processes. Bakó et al. (2021) demonstrated that high-

resolution remote sensing will be significantly helpful in regard to monitoring natural habitats and conserving 

natural resources. Maurya and Kumar (2024) also demonstrated how geospatial data could play an important 

role in making rational long-term decisions based on sustainable development. Given accelerating climate 

change, biodiversity loss, and intensifying anthropogenic pressures, spatial data have become critical for 

monitoring environmental trends, evaluating the condition of our natural resources, and evaluating the impacts 

of conservation and restoration actions. ALS has taken the limelight as an increasingly important remote sensing 

technology for collecting large-scale three-dimensional datasets across a variety of terrestrial landscape types 

such as forests, wetlands, river systems, coastal zones, and mountainous regions. ALS PCs are particularly 

relevant for environmental applications because they provide accurate terrain models and object-level 

information essential for topographic analysis and feature separation (Kraus and Pfeifer, 1998). In addition, they 

enable detailed assessment of vegetation structure, height, spatial distribution, and canopy morphology, which 

are critical for ecological monitoring and habitat characterization (Roussel et al., 2020). It is generally 

acknowledged that ALS offers a superior method for collecting environmental data at the large scale compared 

92



with traditional ground-based surveys, as ALS is easier, less invasive, and spatially continuous, and aligns with 

sustainability-oriented initiatives that measure environmental impacts with the least disturbance to the 

ecosystem, with volumetric sustainability with minimal resources (Figure 1). In spite of these strengths, PC are 

not without their limitations. The spatial resolution and point density of ALS surveys are often limited by sensor 

specifications, flight design, budget, and environmental conditions. Higher density PCs are not sufficiently 

spatially rich to meet detailed modelling purposes, especially in heterogeneous areas with dense vegetation. 

Derivative products of ALS data (e.g., digital terrain models, canopy height models, vegetation indices) are 

important data inputs for ecological assessments, hydrological modelling, and land management purposes, but 

the relative lack of point density often creates barriers to the ability to produce these products. Moreover, it is 

often infeasible to get around these limitations through further data capture because of high operational costs 

and environmental effects of additional flight acquisition (Salvoni et al., 2021). PU provides a computational 

means of enhancing the spatial richness of ALS data without requiring further data capture. PU can result in 

synthetic densifications of PCs that maintain geometric fidelity and topological consistency of the original PC. 

By upsampling existing ALS datasets, we can enhance their applicability. This method creates opportunities for 

data reuse and enrichment, develops work efficiencies, and decreases repeated interaction with the 

environment, which is a sustainable way of working. DL-based workflows and methods will help in this area 

because they can capture fine-scale spatial patterns while preserving surface characteristics in complex terrain, 

which is now increasingly available and applied in Global Navigation Satellite System (GNSS) spatial 

techniques. Informed implications for downstream model development are made from enriched datasets, which 

can help downstream models better predict vegetation structure, provide improved assessments and resource 

allocations, assess change over time, and respond and implement mitigating strategies. Zhang et al. (2022) 

state that most of the existing methods for PU have been designed and evaluated on synthetically or object-

based datasets, typically generated from either indoor scans or computer-aided design (CAD) models, where 

variability in the environments was limited. Kwon et al. (2023) also note that there remains limited research on 

applying these techniques to ALS datasets and to data in practical scenarios, when faced with the issues of 

sensor noise, complex structures, and variability in the environment. Further, little is known about the potential 

for PU to support sustainability goals that include long-term monitoring of forest conditions, watershed 

management, and planning for conservation. This study attempts to fill this void through research that combines 

the PU with actual environmental spatial workflows and places a value on providing enhanced geospatial 

information for sustainable environmental resource management. The work is not only intended to advance our 

technical development of PU methods, but also to promote a more general discussion regarding how 

computational technologies can extend the lifetime and utility of remotely sensed data for the purposes of 

environmental stewardship. 

 

Figure 1: Conceptual workflow of ALS point cloud upsampling from data acquisition to derived products and 

decision support (Source: own creation) 

93



3. Approach to airborne LiDAR upsampling 

The proposed methodological framework for PU aims to mitigate the shortcomings of low-density Light Detection 

and Ranging (LiDAR) acquisitions by creating enriched datasets that enhance the capabilities for environmental 

monitoring, resource management, and planning for sustainability over time. This framework utilizes a modular 

design that permits the inclusion of various data types, methods of algorithms, and methods of computation, 

while maintaining a logical and reproducible order, which is beneficial for the integrity of the workflow. This 

modular design will allow for the adaptation of the framework to a multitude of applications, geographic locations, 

and technical limitations, while allowing continued processing and re-use of sampling whenever appropriate, 

presenting both adaptability and scalability. 

The ALS data collection process begins with the acquisition of the PCs, which are typically collected through a 

national mapping agency, an airborne survey focused on research purposes, or publicly available from various 

governmental and scientific institutions. The data is most often stored as LAS or LAZ files because of the 

compatibility with numerous LiDAR processing tools, although PLY and similar 3D formats are also available, 

particularly for research purposes. The point density of the PCs is also variable depending on flight altitude, 

pulse repetition frequency, scanning angle, and other limitations due to ambient conditions like vegetation cover 

or atmospheric conditions. In a lot of cases, the native resolution may be suitable for large-scale mapping but 

not for detailed assessments that need to have fine-scale representations of terrain morphology, canopy 

architecture, or human-used features.  

Processing is a key step for ensuring that the subsequent PU methods are based on clean and consistent input 

data. Processing generally involves identifying and eliminating noise and statistical outliers due to sensor error, 

atmospheric effects, or incorrect classifications in the initial ground filtering. Elevation normalization is applied 

to convert raw height into a consistent vertical reference system, typically using geoid or ellipsoid corrections to 

ensure consistency when integrating data from multiple sources. Ancillary properties - such as return intensity, 

classifications (land use, ground cover, etc.), and RGB colour values - should also be preserved and, where 

applicable, normalized or resampled so that they can be propagated through the same PU process, undistorted 

or misaligned. Proportionately preparing ancillary properties will also allow access to more advanced modelling 

procedures and frameworks, which might use both geometry and radiometric information. 

The PU process is the central focus of the framework, and it can be executed in many different methodological 

families. Traditional methods such as triangulated irregular network-based (TIN), inverse distance weighting 

(IDW), and kriging methods are user-friendly and computationally inexpensive. However, they are subject to 

oversmoothing and degradation of detailed structural information. Optimization-based methods attempt to 

enhance spatial uniformity, stabilize structural elements, or both by exploring the problem through the lens of 

energy minimization. In recent years, DL-based approaches to PU have presented a fruitful direction by 

employing deep neural architectures to learn the geometry structure and fine details. The first end-to-end 

framework to learn multi-scale point features directly into densely sampled point sets from sparse inputs is PU-

Net (Yu et al., 2018). They model a point-wise neural network architecture with a focus on multi-scale features 

to create uniform point distributions. MPU (Yifan et al., 2019) next conceptualizes a patch-based progressive 

upsampling framework that will learn to reconstruct fine-grained geometric structure through progressive levels 

of detail. Subsequently, PU-GAN (Li et al., 2019) conceptualizes an adversarial framework that improves point 

distributions to uniformity and surface proximity regardless of the input density or perturbation. Lastly, PUGeo-

Net (Qian et al., 2020) uses a neural network to integrate discrete differential geometry with the upsampling 

process by jointly learning point coordinates and normals and learning to represent intrinsic surface properties. 

These methods for ALS data require minor adjustments to the DL method and parameters to appropriately 

address the irregular point density in the affine density in the vegetated areas, the partial penetration of laser 

pulses through canopy layers, and the differences in scale between open areas and heavily forested areas. 

The post-processing stage is where the output from the upsampling stage is improved to confirm that the 

enriched dataset is an accurate portrayal (in terms of location) and is of sufficient quality for its intended purpose. 

This involves a second round of outlier removal to eliminate artefacts caused by the densification process, 

implementation of uniform spacing of points to manage clustering, and the merging of deltas with the original 

dataset such that both datasets remain intact. It is possible that this step involves reclassification of points based 

on their location in relation to a terrain model or canopy model, creating thematic consistency across the dataset. 

The advantage of these enhanced PCs is the generation of higher-quality environmental products. DTMs can 

be rebuilt with better representation of small-scale topographic features, i.e., micro channels, embankments, or 

small-scale erosion features. CHMs will have increased sampling of upper and lower canopy layers, allowing 

for more accurate estimation of vegetation structure and biomass. From these outputs (DTM, CHM, vegetation 

indices, and improved land cover classifications), hydrologic parameters will be more robust for watershed 

analysis as well. Finally, these products will be the inputs to downstream applications ranging from biodiversity 

assessments to risk analysis of infrastructure. 

94



It is important to incorporate validation and quality evaluation to ensure the credibility of upsampled products. 

For example, geometric accuracy can be measured by RMSE, Chamfer distance, and Hausdorff distance, while 

thematic accuracy may be evaluated by the comparison of derived DTMs, CHMs, or classification maps to 

trusted reference data. The validation is designed to be hierarchical, where high-density airborne or Unmanned 

Aerial Vehicle (UAV)-borne lidar, terrestrial laser scanning, or terrestrial surveys provide points of reference and 

are used quantitatively and statistically to provide estimates of performance both spatially and statistically. The 

assessment of performance through validation and quality evaluations can help identify systematic bias, 

localized errors, or conditions where some PU methods failed to perform adequately. 

The framework is intended to serve as part of an operational workflow that can support forestry management, 

watershed modelling, protection of coastlines, and planning for conservation. By embedding cloud-based 

processing solutions, enabling batch processing, and eventually nearly real-time processing, the costs and 

environmental footprint associated with aerial surveys can be reduced by facilitating the use of purpose-built 

aerial surveys. More importantly, we provide a concrete and flexible framework for upsized PCs as resolution 

and needs for environmental science and policy continue to evolve. 

4. Challenges and limitations 

Though  PU clearly has advantages for improving ALS datasets, several limitations remain. It is not clear how 

models will transfer between different landscapes, as a given model trained on one biome does not appropriately 

generalize to other biomes that differ significantly in terms of canopy structure or the morphologies of the ground 

and canopy. Additionally, ALS acquisitions are notoriously heterogeneous (i.e., flight altitude, scan angle, and 

sensor specifications), which will lead to differences in how PU workflows are standardized. Moreover, the 

computation cost of prominent DL methods is notable, and if large-scale or near-real-time monitoring tasks are 

to be conducted without the use of extensive hardware resources, these methods remain challenging to 

implement. Another important limitation is the type of artefacts/biases that arise during the PU process: while 

the PC is being densified, there is also the potential to generate artificial structural patterns that can misrepresent 

true surface properties - this is especially true for vegetated surfaces where the effects of occlusion predominate. 

As such, it is important to consider model selection, benchmarking against independent reference datasets, and 

possibly hybrid methods that utilize deep learning with physics-based or statistical models to improve 

generalisability and or reproducibility. 

5. Conclusion and future directions 

Further model generalization is necessary, reducing computational requirements, and ensuring that enriched 

PCs retain geometric and attribute information. Future development work will likely see, but not be limited to, 

DL models that preserve attributes, multi-sensor inputs including UAV-based LiDAR and multispectral imagery, 

and cloud-based processing pipelines enabling near-real-time environmental condition monitoring. Additional 

work in any of these areas can transition PU from research projects into operational activity as part of sustainable 

environmental management. In practical usage, the conceptual framework outlined here demonstrates how PU 

is applied to enhance existing environmental monitoring programs. For the forestry sector, denser PCs provide 

more accurate estimates of tree- and stand-level biomass estimates, as well as contribute to more accurate 

large-scale forest inventory programs. For watershed hydrology, denser PCs enable the modelling of river 

networks, floodplains, and slopes susceptible to erosion at a finer resolution, enhancing hydrological 

assessments. Greater spatial detail in coastal environments is beneficial in monitoring shoreline change and 

sediment processes to support erosion controls and climate adaptation. Similarly, denser PCs enhance 

conservation planning by allowing for enhanced detection, monitoring, and tracking of habitat structure, species 

niche space functionality, and biodiversity indicators, which may enhance stewardship and management. 

While this synthesis does not present empirical results, it highlights the potential to utilize workflows for PU and 

to reduce expensive re-surveys of the same area, as well as increasing confidence in the downstream ecological 

models. This conceptual study demonstrates that the upsampling of PC data increases spatial detail while 

incurring potential cost savings for the sustainability of monitoring environmental conditions. An important 

contribution of the research presented above is to show evidence of a shift to an operational workflow for 

resource management that relies on data and incorporates a more resilient approach. 

Moving forward, work to calculate and validate PU workloads and to test different methods for different dataset 

needs must continue, especially to determine which methods can improve the precision and accuracy of 

different ecological indicators. The development of practices for the standardization of PU methods for clear 

long-term monitoring and sustainability planning should also be a consideration for future work. In this way, PU 

will evolve from a new and promising concept to a routine, valuable method for addressing the environmental 

and societal dilemmas facing our coupled societies in the next few decades. 

95



References 

Bakó G., Molnár Z., Bakk L., Horváth F., Fehér L., Ábrám Ö., Morvai E., Biro C., Pápay G., Fűrész A., Penksza 

K., Pácsonyi D., Demény K., Juhász E., Dékány D., Csernyava L., Illés G., Molnár A., 2021, Toward a high 

spatial resolution aerial monitoring network for nature conservation — How can remote sensing help protect 

natural areas? Sustainability, 13(16), DOI:10.3390/su13168807. 

Hollaus M., Wagner W., Maier B., Schadauer K., 2007, Airborne laser scanning of forest stem volume in a 

mountainous environment. Sensors, 7(8), 1559–1577, DOI:10.3390/s7081559. 

Kraus K., Pfeifer N., 1998, Determination of terrain models in wooded areas with airborne laser scanner data. 

ISPRS Journal of Photogrammetry and Remote Sensing, 53(4), 193–203, DOI:10.1016/S0924-

2716(98)00009-4. 

Kwon S., Hur J.-H., Kim H., 2023, Deep learning-based point cloud upsampling: a review of recent trends. JMST 

Advances, 5, 105–111, DOI:10.1007/s42791-023-00058-6. 

Li R., Li X., Fu C., Heng D.C.P., 2019, PU-GAN: A point cloud upsampling adversarial network. Proceedings of 

the IEEE International Conference on Computer Vision, 1, 7202–7211, DOI:10.1109/ICCV.2019.00730. 

Maurya A.K., Kumar A., 2024, The role of GIS in the study of sustainable development and environmental 

management. International Journal for Multidisciplinary Research, 6, 1–13, DOI: 

10.36948/ijfmr.2024.v06i06.32879. 

Qian Y., Hou J., Kwong S., He Y., 2020, PUGeo-Net: A geometry-centric network for 3D point cloud upsampling. 

In: Vedaldi A., Bischof H., Brox T., Frahm J.-M. (Eds.), Computer Vision – ECCV 2020, Springer, Cham, 

Switzerland, 752–769, DOI:10.48550/arXiv.2002.10277. 

Roussel J.-R., Auty D., Coops N.C., Tompalski P., Goodbody T.R.H., Meador A.S., Bourdon J.-F., de Boissieu 

F., Achim A., 2020, lidR: An R package for analysis of airborne laser scanning (ALS) data. Remote Sensing 

of Environment, 251, 112061, DOI:10.1016/j.rse.2020.112061. 

Salvoni D., Boselli A., Sannino A., Parlato L., Ejrnaes M., Zhang C., You L., Wang X., Amoruso S., Pepe G.P., 

2021, Demonstration of atmospheric lidar measurement in the infrared wavelength domain with a 

superconducting nanowire single photon detector. Chemical Engineering Transactions, 84, 175–180, 

DOI:10.3303/CET2184030. 

Sha C., Lu Z., Li R., Zheng H., Duodu E.A., 2016, A point cloud registration algorithm for free-form surface. 

Chemical Engineering Transactions, 51, 469–474, DOI:10.3303/CET1651079. 

Skidmore A.K., Pettorelli N., Coops N.C., Geller G.N., Hansen M., Lucas R., Mücher C.A., O’Connor B., Paganini 

M., Pereira H.M., Schaepman M.E., Turner W., Wang T., Wegmann M., 2015, Environmental science: Agree 

on biodiversity metrics to track from space. Nature, 523(7561), 403–405, DOI:10.1038/523403a. 

Venter O., Sanderson E.W., Magrach A., Allan J.R., Beher J., Jones K.R., Possingham H.P., Laurance W.F., 

Wood P., Fekete B.M., Levy M.A., Watson J.E.M., 2016, Sixteen years of change in the global terrestrial 

human footprint and implications for biodiversity conservation. Nature Communications, 7, 1–11, 

DOI:10.1038/ncomms12558. 

White J.C., Coops N.C., Wulder M.A., Vastaranta M., Hilker T., Tompalski P., 2016, Remote sensing 

technologies for enhancing forest inventories: A review. Canadian Journal of Remote Sensing, 42(5), 619–

641, DOI:10.1080/07038992.2016.1207484. 

Yifan W., Wu S., Huang H., Cohen-Or D., Sorkine-Hornung O., 2019, Patch-based progressive 3D point set 

upsampling. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019, 

5951–5960, DOI:10.1109/CVPR.2019.00611. 

Yu L., Fu X.L.C., Heng D.C.P., 2018, PU-Net: Point cloud upsampling network. Proceedings of the IEEE 

Conference on Computer Vision and Pattern Recognition, 2790–2799, DOI:10.1109/CVPR.2018.00295. 

Zhang T., Filin S., 2022, Deep-learning-based point cloud upsampling of natural entities and scenes. The 

International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XLIII-B2-

2, 321–327, DOI:10.5194/isprs-archives-XLIII-B2-2022-321-2022. 

Zhang Y., Zhao W., Sun B., Zhang Y., Wen W., 2022, Point cloud upsampling algorithm: A systematic review. 

Algorithms, 15(4), DOI:10.3390/a15040124. 

96


	COS25_0085.pdf
	Potential of Point Cloud Upsampling for Environmental Protection: Enhancing Airborne LiDAR Data for Sustainable Resource Management