UAV Image-Based 3D Reconstruction Technology in Landslide Disasters: A Review

Abstract

Global geological conditions are complex and variable, characterized by frequent plate movements, earthquakes, and volcanic eruptions. Coupled with significant climate differences, various factors interact to trigger frequent landslide disasters, resulting in substantial losses of life and property. Therefore, landslide monitoring is crucial. Traditional monitoring technologies face limitations when dealing with complex terrains and meeting the demands for high timeliness, while unmanned aerial vehicles (UAVs), with their maneuverability, high resolution, and ability to operate in hazardous environments, have been widely applied in landslide monitoring. This paper provides a comprehensive review of UAV-based 3D reconstruction for landslides, detailing the characteristics and application cases of UAVs, explaining the functions and limitations of sensors such as optical sensors and light detection and ranging (LiDAR), and exploring 3D reconstruction methods based on UAV imagery, LiDAR, and hybrid approaches. It analyzes the applications of UAV 3D reconstruction in landslide emergency investigation, monitoring, and disaster assessment. The paper identifies the technical challenges faced in these applications and proposes corresponding solutions. In addition, UAV-based 3D reconstruction technology—with its centimeter-level spatial resolution—enables the precise delineation of landslide extent and hazard potential, thereby enhancing monitoring accuracy and improving the efficiency of emergency investigations. This technology provides strong technical support for landslide research and prevention, with significant implications for reducing landslide disaster losses.

1. Introduction

Global geological conditions are complex and variable, with frequent plate activities, as well as geological events such as earthquakes [1,2,3] and volcanic eruptions [4,5,6], occurring frequently. Coupled with significant climatic differences, extreme weather conditions such as heavy rainfall [7,8,9,10] and hurricanes [11,12,13] are not uncommon. The interaction of these factors frequently triggers landslide disasters [14,15,16]. A landslide refers to the phenomenon in which surface soil and rock masses move downslope under the influence of gravity [17]. When the shear stress on a slope exceeds its shear strength, it typically triggers a landslide, which may be induced by intense short-duration or prolonged rainfall, earthquakes, or human activities [18,19,20,21]. Landslide disasters are among the most frequent, widely distributed, and damaging geological hazards worldwide, posing a serious threat to human life, property, and major engineering facilities [22]. Consequently, research into landslide phenomena holds substantial social and economic significance [23].

Traditional slope inclinometers measure the magnitude and direction of slope movements, whereas the Global Navigation Satellite System (GNSS) continuously monitors the surface displacements of landslide masses. However, each instrument only monitors a single point, making it necessary to integrate remote sensing techniques to capture the entire landslide body [24,25]. Remote sensing technologies such as optical imagery and airborne LiDAR are widely used for large-scale landslide monitoring [26]. Compared with slope inclinometers and GNSS, remote sensing can acquire spatially extensive data in a single survey, improving efficiency. However, limitations in temporal and spatial resolution hinder its application in complex terrain, at large scales, and under time-sensitive requirements [26,27]. In particular, UAVs offer high resolution at a low cost and can safely acquire data in hazardous environments, making them an important tool for obtaining spatial information [28]. Furthermore, breakthrough advances over the past three years—including multisensor fusion [29], AI-assisted reconstruction algorithms [30], and vertical takeoff and landing (VTOL) hybrid-wing UAVs that overcome endurance limitations in complex terrain [31]—have further expanded UAV operational capabilities in challenging terrain and time-sensitive scenarios [32,33]. The emergence of UAVs has significantly advanced landslide research, particularly in field investigations [34,35]. To characterize the spatial and temporal distribution and morphological features of landslides, 3D reconstruction is essential [36]. Real-scene 3D modeling enables the construction of high-resolution models of both hazard-inducing and hazard-bearing bodies in high-risk landslide areas, facilitating the acquisition of comprehensive, real-time spatial and attribute information for these regions. This advancement strongly promotes technological progress in landslide risk prevention and control and enhances disaster prevention, mitigation, and emergency response capabilities [15,16,37].

UAV 3D reconstruction technology plays an irreplaceable role in landslide research, supporting all stages of landslide evolution and providing comprehensive support for landslide investigation, monitoring, and management from multiple dimensions. With its unique flexibility, high precision, and efficiency, it supports landslide research and practice in all aspects, from investigation to monitoring and disaster assessment, providing strong technical support and a reliable basis for decision-making in the prevention, response, and recovery of landslide disasters.

In summary, UAV-based 3D reconstruction technology serves as a vital tool for landslide investigation, monitoring, and disaster assessment. Its flexibility, high precision, and efficiency provide robust technical support and a reliable foundation for decision-making in disaster prevention, emergency response, and post-disaster recovery.

By reviewing and synthesizing the relevant literature, this paper comprehensively explores the application of UAV 3D reconstruction technology in landslide research. To conduct this review, papers related to UAV 3D reconstruction and landslides were retrieved from Web of Science without temporal restrictions, ultimately including 143 references spanning from 1999 to 2025. Figure 1 shows the temporal distribution of the references, with literature from 2020 to 2025 accounting for 71%, indicating a rapid growth trend in this research field. Earlier studies primarily focused on fundamental descriptions and theoretical explanations of related algorithms, thereby laying the foundation for subsequent applied research. Figure 2 presents the distribution of the references across major journals, with Remote Sensing and Landslides being particularly prominent, which account for 13.3% and 7%, respectively. These two journals are also among the most authoritative core journals in this field. Additionally, the research covered various terrain environments, including mountainous areas and forest-covered regions, as well as multiple disaster types such as rainfall-induced landslides and earthquake-triggered landslides, fully demonstrating the regional adaptability and diversity of UAV-based applications. Furthermore, the literature cases were distributed across landslide-prone countries such as China, India, the United States, Italy, and Switzerland, reflecting both pronounced regional characteristics and global application value, providing an important basis for technology promotion in different geographical contexts. This review aims to provide meaningful updates to landslide monitoring research and to suggest directions for addressing related challenges.

2. UAV-Based Landslide Monitoring

2.1. UAV Types for Landslide Monitoring

UAVs are pilotless aircraft operated by radio remote control equipment or self-contained program control devices. UAVs come in various shapes and specifications, with flight altitudes ranging from a few meters to several kilometers, and weights varying from a few kilograms to hundreds of kilograms. According to their structural configuration and takeoff/landing characteristics, UAVs can be categorized into four main types: multirotor UAVs, fixed-wing UAVs, VTOL aircraft, and airships [38]. In landslide monitoring, multirotor UAVs [39,40,41,42,43,44,45] and fixed-wing UAVs [46,47,48,49,50,51,52] are the most commonly used. Table 1 provides a comparison of the typical flight parameters of fixed-wing and multirotor UAVs.

2.1.1. Multirotor UAVs

Multirotor UAVs are unmanned aircraft that generate lift by rotating multiple rotors [39]. Due to their outstanding vertical takeoff and landing capabilities, hovering abilities, and capacity to acquire high-precision data [53], multirotor UAVs are particularly suitable for detailed 3D modeling and real-time monitoring in landslide areas, which has led to their extensive application in landslide research.

For instance, Lucieer et al. [18] employed an OktoKopter remote-controlled UAV, equipped with a digital camera and a Global Positioning System (GPS) receiver, to perform a 3D surface reconstruction of the Home Hill landslide in Southern Tasmania, Australia. By applying image correlation algorithms, they achieved the accurate mapping of landslide displacement. Similarly, Tanteri et al. [54] utilized the Saturn multirotor UAV, developed by the Department of Earth Sciences of Florence, to construct 3D models and conduct long-term dynamic monitoring of two landslide areas in Ricasoli village, Italy, thereby providing time-series data to support geological evolution analyses. The adaptability of multirotor UAVs becomes particularly evident in complex terrain. Ren et al. [17] leveraged DJI Phantom 4 Pro to conduct a full-area 3D reconstruction of the Zongling landslide in a coal mining district of Liupanshui, Guizhou, obtaining a complete terrain model that captured fine fissure features. Meanwhile, Teo et al. [55] utilized DJI Phantom 4 RTK to extract the 3D displacement field of the large-scale Wufeng landslide in Hualien, Taiwan, where its centimeter-level positioning accuracy effectively captured the dynamic deformation characteristics of the landslide body. These cases collectively demonstrate that multirotor UAVs, owing to their modular payload configurations and high operational maneuverability, are emerging as a core technological tool for fine-scale landslide monitoring and post-disaster emergency mapping.

2.1.2. Fixed-Wing UAVs

Fixed-wing UAVs are unmanned aircraft that generate lift using fixed wings, which remain stationary throughout flight. Compared to multirotor UAVs, fixed-wing UAVs have stronger endurance, faster flight speeds, and broader coverage capabilities [56], making them particularly advantageous in large-scale landslide surveys and mapping [27].

Rau et al. [46] utilized a fixed-wing UAV equipped with a consumer-grade digital camera, a Garmin GPS receiver, and an Attitude and Heading Reference System (AHRS) to monitor hundreds of landslides triggered by Typhoon Morakot in Taiwan. This work not only validated the feasibility of landslide monitoring algorithms but also highlighted the technical suitability of fixed-wing UAVs for large-scale rapid disaster mapping, as their endurance is sufficient for the continuous data acquisition required over large areas post-typhoon. Similarly, Anders et al. [47] employed a MAVinci Sirius 1 fixed-wing UAV to construct digital surface models (DSMs) of a representative shallow landslide within an agricultural catchment in Pamplona, Northern Spain. By comparing multitemporal datasets, they confirmed the practical utility of fixed-wing platforms in detailed geomorphological studies, particularly in landslide monitoring scenarios requiring long-term topographical evolution tracking. Furthermore, in a study conducted in the West Black Sea region of Turkey, Comert et al. [22] acquired aerial imagery of a landslide area using a fixed-wing UAV system and achieved 80% accuracy in semi-automated landslide identification through an object-based image analysis approach. These studies collectively demonstrate that fixed-wing UAVs possess an indispensable technological role in landslide research requiring a balance between large-area investigation and long-term monitoring.

In landslide disaster research, fixed-wing UAVs (e.g., Hi-Target V100) are particularly suitable for large-scale investigations and rapid emergency mapping, benefiting from their 150-min endurance and 30 m/s cruise speed. In contrast, multirotor UAVs (e.g., DJI Matrice 350) are indispensable for the fine-scale modeling of small, hazardous areas, owing to their VTOL capability, hovering stability, and high-precision data acquisition. However, current technology faces three major bottlenecks: the limited battery energy density of multirotor UAVs hinders long flight times, while integrating both LiDAR and optical payloads hinders long flight times when integrating both LiDAR and optical payloads; the imaging quality of optical sensors degrades significantly in rainy and foggy conditions, while LiDAR’s accuracy is reduced by atmospheric particulates; and visual line-of-sight (VLOS) regulations in Europe and America restrict the long-range operational capabilities of fixed-wing UAVs in complex terrain.

Future efforts should focus on enhancing hardware performance by developing high-energy-density batteries (such as solid-state batteries) and lightweight composite materials; mitigating single-sensor limitations through AI-driven fusion algorithms for optical and LiDAR point clouds; and establishing an air–ground collaborative monitoring framework that integrates “fixed-wing UAVs for global scanning + multirotor UAVs for local detailed surveying + ground-based GNSS/radar”. As sensor miniaturization and autonomous navigation technologies mature, VTOL hybrid-wing UAVs—which combine the flexibility of vertical takeoff with the long endurance of fixed-wing aircraft—are poised to become an ideal solution. Coupled with 5G real-time data transmission and edge computing, these UAV platforms are expected to enable a full-chain intelligent transformation of landslide monitoring toward “high-precision early warning, dynamic response, and intelligent decision-making”, providing more robust technological support for disaster prevention and control.

Table 1. Comparison of flight parameters between fixed-wing and multirotor UAVs.

Fixed-Wing UAV	Multirotor UAV (Light)	Multirotor UAV (Heavy)
Hi-Target V100 [57]	DJI Phantom 4 Pro [58]	DJI Matrice 350 [58]
Maximum Flight Time: 150 min	Maximum Flight Time: 30 min	Maximum Flight Time: 55 min
Maximum Flight Speed: 30 m/s	Maximum Flight Speed: 20 m/s	Maximum Flight Speed: 23 m/s
Maximum Payload: 2 kg	Aircraft Weight: 1.388 kg	Maximum Payload: 2.7 kg
Maximum Flight altitude: 6000 m	Maximum Flight altitude: 6000 m	Maximum Flight altitude: 7000 m
Adaptability to Work Environment: Open Areas	Adaptability to Work Environment: Complex Terrain

Table 1. Comparison of flight parameters between fixed-wing and multirotor UAVs.

Fixed-Wing UAV	Multirotor UAV (Light)	Multirotor UAV (Heavy)
Hi-Target V100 [57]	DJI Phantom 4 Pro [58]	DJI Matrice 350 [58]
Maximum Flight Time: 150 min	Maximum Flight Time: 30 min	Maximum Flight Time: 55 min
Maximum Flight Speed: 30 m/s	Maximum Flight Speed: 20 m/s	Maximum Flight Speed: 23 m/s
Maximum Payload: 2 kg	Aircraft Weight: 1.388 kg	Maximum Payload: 2.7 kg
Maximum Flight altitude: 6000 m	Maximum Flight altitude: 6000 m	Maximum Flight altitude: 7000 m
Adaptability to Work Environment: Open Areas	Adaptability to Work Environment: Complex Terrain

2.2. Categories of UAV-Mounted Monitoring Sensors

Sensors can be classified into active and passive sensors based on whether they need to emit detection signals to the target objects. The rapid development of sensor technology has enabled UAVs to operate and navigate effectively even in complex environments. There are various types of sensors for UAVs, which serve as the “eyes”, “ears”, and “feelers” of the UAV, providing robust environmental perception capabilities. By selecting appropriate sensors for specific application scenarios, specific functions and objectives can be achieved. Commonly used sensors in landslide applications include optical sensors and LiDAR. Optical sensors can acquire high-resolution images of the ground, while LiDAR can obtain large-scale 3D point clouds, which can subsequently be employed to construct detailed 3D models of landslide areas. These 3D models can be utilized for monitoring landslides, guiding emergency rescue, and supporting post-disaster reconstruction.

Table 2. Functions and limitations of UAV sensors.

Visible Light Camera	Thermal Infrared Camera	Multispectral Camera	LiDAR
Main Application: High-resolution color imaging	Main Application: Infrared radiation and temperature detection	Main Application: Multispectral vegetation classification	Main Application: Laser ranging, 3D modeling
Drawbacks: Dependent on daylight conditions	Drawbacks: Lower spatial resolution	Drawbacks: Limited spatial resolution	Drawbacks: Atmospheric interference, high cost
	Specific influencing factors: Image contrast decreases by 40% under rainy conditions, and more than 50% of texture information is lost under foggy conditions.		Specific influencing factors: heavy-fog conditions reduce point cloud density by approximately 30%, while rainy/overcast conditions increase the error by 5–10 cm.

presents the capabilities and limitations of commonly used UAV sensors.

2.2.1. Optical Sensors

Optical sensors are a type of sensor that operate based on optical principles. They have many advantages, such as non-contact and non-destructive measurement, minimal interference, high-speed transmission, and the ability to perform remote measurement and control. Compared to LiDAR, optical sensors are more easily integrated across various platforms [59]. Common optical sensors include visible light cameras [18,60], thermal infrared cameras [41,43,61,62], and multispectral cameras [63,64,65,66,67,68].

Sensors in the visible light range are the most commonly used, typically recording intensity information across three bands of the visible spectrum. Traditional photogrammetry and oblique photogrammetry primarily use visible light cameras. This imaging technology enables the capture of objects’ true colors, textures, and other appearance details while offering the advantages of low cost and operational simplicity [69].

Visible light cameras can be adversely affected by natural conditions such as weather and fog during landslide monitoring, while thermal infrared cameras have strong penetration capabilities and are independent of lighting conditions, making them well-suited for the complex geographical environments and extreme weather conditions associated with landslides. By comparing multiple thermal infrared images over time, precise monitoring of landslide displacement and crack changes can be achieved [70]. After a landslide occurs, UAVs equipped with thermal infrared sensors can quickly generate high-resolution 3D temperature maps remotely. By integrating these data, detailed 3D topographic landslide models can be constructed, helping to analyze movement trends, delineate affected areas, and estimate the volume of displaced material. In emergency situations, this approach enables rapid information acquisition, supports mitigation planning and risk assessment, and can be used for search and rescue operations to locate individuals trapped after a landslide [71].

Multispectral sensors differ from the above two types in that they can simultaneously acquire electromagnetic spectral information across multiple bands and perform detailed analyses to identify material composition and vegetation characteristics [72]. Newly occurred landslides often exhibit distinct features, which can be readily detected using UAVs equipped with multispectral sensors [73]. In addition, these sensors can also generate orthophotos and 3D terrain models, as well as identify and extract landslide attributes such as location, impact area, and displaced material volume [74]. For example, the DJI Phantom 4 multispectral UAV can acquire data across visible light, near-infrared, and short-wave infrared bands. These data can be used to extract vegetation indices (VI), particularly the Normalized Difference Vegetation Index (NDVI), which can serve as an indicator for landslide hazard assessment [69].

2.2.2. LiDAR Sensors

LiDAR, or light detection and ranging, is an active optical remote sensing technology renowned for its high precision in geometric data acquisition [75]. LiDAR technology serves as a key method for acquiring high-resolution digital terrain data, and with the miniaturization of LiDAR equipment and the enhancement of UAV payload capabilities, UAV-mounted LiDAR has become increasingly common [27]. Owing to the intrinsic properties of LiDAR, when combined with UAV platforms, data can be collected more efficiently and with reduced dependency on terrain conditions. UAVs equipped with LiDAR can penetrate vegetation, thereby eliminating vegetation interference and establishing high-precision 3D models, which greatly facilitates landslide monitoring and subsequent disaster assessment [52]. Currently, commercial UAV LiDAR systems are evolving towards lighter weights and lower costs. For example, the DJI Zenmuse L2 integrates an RGB camera with a Livox Mid-40 laser LiDAR, weighing only 0.9 kg and achieving a point rate of 240,000 pts/s, while supporting real-time point cloud coloring (DJI(DJI Innovation Technology Co., Ltd., Shenzhen, China), 2023 [76]). The Hesai PandarXT-32 (Hesai Technology Co., Ltd., Shanghai, China), utilizing MEMS technology, achieves centimeter-level accuracy (Hesai, 2024 [77]) and has reduced its price to approximately 30% of that of traditional systems, thereby facilitating the large-scale deployment of UAV-based landslide monitoring.

However, LiDAR also exhibits certain limitations: it is sensitive to atmospheric factors such as fog, rain, and dust, which can interfere with its operation and compromise data acquisition accuracy [78]. While LiDAR can obtain precise 3D models, its complex operation and high costs hinder its application in some areas, particularly for repeated measurements [79]. Nevertheless, some current commercial LiDAR systems based on UAVs offer enhanced measurement precision and data-processing efficiency, as well as broad development prospects, such as the DJI Zenmuse L2 and Pegasus D2000. These systems provide greater convenience and possibilities for landslide-related investigations, monitoring, and assessments, attracting significant attention in the field, and they are expected to further advance the application of airborne LiDAR in landslide studies.

By equipping UAVs with optical, LiDAR, and other sensors, researchers now have access to a broader array of data acquisition methods and diverse types of deliverables for landslide studies. UAVs can rapidly generate high-precision 3D models of landslide areas, making significant contributions to landslide research. Notably, recent technological breakthroughs have substantially reduced the weight and size of UAV-mounted sensors, thereby improving equipment portability and enabling the integrated deployment of multiple sensor payloads. For example, the DJI Zenmuse L2 achieves an integrated design of LiDAR and an RGB camera, reflecting an industry trend toward highly integrated measurement tools. This innovation allows for the simultaneous acquisition of 3D point cloud data, RGB imagery suitable for photogrammetric processing, and potentially hyperspectral, multispectral, or thermal infrared datasets within a single UAV flight. Such advancements are expected to significantly enhance the efficiency and accuracy of landslide assessment.

To clearly define the scope of the application of UAV-based 3D reconstruction technology in landslide disaster research, this study proposes a comprehensive technological framework, as illustrated in Figure 3. This map establishes a direct link between specific landslide scenarios, optimal technical configurations, and targeted application outcomes, providing a decision-making framework for researchers and practitioners.

3. UAV-Based 3D Reconstruction Techniques

3.1. Three-Dimensional Reconstruction Based on Image

The 3D reconstruction method based on UAV imagery allows for access to diverse locations and enables data acquisition at varying ground resolutions by adjusting the flight altitude [80]. This enables rapid landslide identification and parameter extraction from multiple dimensions and perspectives, thereby overcoming the limitations of two-dimensional imagery. Moreover, it provides essential technical and data support for both landslide management and subsequent research [81]. UAV oblique photogrammetry employs multiple sensors mounted on UAVs to simultaneously capture image data from one vertical and four oblique angles, thus enabling the acquisition of high-precision texture and geometric information of ground objects after processing [82]. Oblique photogrammetry is cost-effective, efficient, and capable of producing metrically reliable images, making it the current mainstream approach for 3D reconstruction. Figure 4 and Figure 5 illustrate the principles of oblique photogrammetry and the workflow of 3D reconstruction based on oblique photogrammetry, respectively.

Three-dimensional reconstruction based on UAV imagery primarily consists of the following key steps: First, operators design detailed flight routes and image acquisition points according to the mapping area [83]. The UAV then takes off automatically along the planned route, captures images at designated locations, and simultaneously records GPS coordinates and camera pose information. After completing the image acquisition, the UAV returns automatically. Second, Scale-Invariant Feature Transform (SIFT) or Speeded-Up Robust Feature (SURF) algorithms are applied to detect and register feature points across multiple images, yielding scale- and rotation-invariant features. Subsequently, bundle adjustment and triangulation are employed to construct error equations, followed by iterative optimization to determine the 3D coordinates of spatial feature points and camera poses. Next, a multiview stereo (MVS) matching strategy is adopted to exploit image information comprehensively, enabling the reconstruction of a dense 3D point cloud. Then, based on the topology of the point cloud, Delaunay triangulation is used to construct an unstructured mesh. Finally, the mesh is converted into a 3D solid model, and textures are applied to fit the original image textures, thereby generating a photorealistic 3D scene model [82]. Each step complements the others, enabling the transformation from images to a 3D model. Among these processes, UAV flight path planning, feature extraction, and image matching have consistently been research hotspots.

In order to build an optimal 3D model, an initially appealing idea is to capture as many images as possible. However, due to the limitations of UAV endurance and cost, as well as the fact that an excessive number of images may adversely affect the reconstruction accuracy of specific points, this approach is neither efficient nor applicable in many real-world scenarios [84]. Another method is to select high-quality images from the captured dataset for 3D reconstruction, thereby mitigating excessive computation and minimizing the impact of low-quality images on the model. Although this approach reduces computational overhead and image interference by filtering a subset of images, it still suffers from drawbacks such as long data acquisition times, high computational costs for filtering and processing the remaining images, unoptimized camera rotation parameters, and reliance on the complete image set [85,86]. The most effective solution is to plan the UAV flight path prior to data acquisition, which constitutes a central topic in viewpoint and path planning research across multiple domains.

According to whether a prior environment model is required, UAV path planning methods can be divided into model-free methods and model-based methods.

Most model-free methods adopt the “Next Best View” (NBV) approach. The core logic of this method is to iteratively determine the optimal solution based on local information; that is, by continuously evaluating unknown regions within the current environment map and dynamically selecting the observation position that maximizes information gain. Model-free methods mainly include boundary-based methods [87,88], volumetric planning [89,90], and surface-based planning [91,92]. Boundary-based methods can quickly explore unknown areas but lack the ability to accurately model complex 3D surfaces. Volumetric planning can generate more accurate and complete voxel models, but it faces challenges in reconstructing intricate 3D surfaces. Surface-based planning can fully reconstruct high-quality 3D surface models, but it requires substantial memory and computational resources. Therefore, in practical applications, the choice of path planning method should be adapted to the scanning objectives and available computational capabilities.

Model-based methods utilize a rough representation of scene geometry for path planning. Agisoft Metashape version 1.5 integrates a model-based image acquisition path planning function [85] by constructing a set of viewpoints with sufficient overlap. However, the static network of camera poses generated by this technique has two limitations: first, it cannot guarantee information completeness in actual data acquisition scenarios; second, it does not explicitly incorporate the accuracy optimization objective of the final point cloud. The recently proposed “Explore-Then-Exploit” (ETE) method [85,93] divides UAV mission planning into two stages: (i) in the exploration stage, an initial set of viewpoints and a flight trajectory are generated using existing planners or manual planning approaches [94]; (ii) in the exploitation stage, the viewpoint layout and trajectory design are optimized with the assistance of a geometric proxy model to obtain image data suitable for 3D reconstruction [95,96].

Model-free methods rely on the NBV strategy to achieve rapid exploration without prior knowledge of the scene, but their nonlinear, local-search mechanisms are prone to being trapped in local optima, leading to the incomplete coverage of object details. Moreover, their reliance on short-range acquisition modes employing stereo or depth cameras leads to efficiency bottlenecks in large-scale 3D reconstruction. Although the ETE method leverages geometric priors to achieve global coverage and accuracy optimization, current offline planning approaches lack real-time feedback mechanisms during data acquisition. Future research may focus on developing online pose optimization strategies that integrate onboard computing with image feedback, enhancing computational capacity through 5G connectivity and cloud-based collaboration and advancing lightweight 3D data processing techniques to eliminate the need for secondary flights.

In the field of computer vision, related technologies such as feature extraction and image matching theory [97], Structure-from-Motion (SfM) algorithms, and multiview stereo (MVS) algorithms are continuously evolving [98]. The rapid development of these technologies serves as a powerful catalyst, driving substantial advancements in 3D reconstruction techniques derived from two-dimensional imagery.

Feature extraction and image matching are fundamental components of digital photogrammetry. Feature extraction involves identifying representative elements within images that capture the shape, texture, and spatial relationships of objects [99]. Image matching is the process of extracting homologous points from multiple overlapping images using algorithms [100]. Feature extraction serves as the foundation for image matching. Image matching methods can be divided into intensity-based and feature-based approaches. Although intensity-based image matching was introduced earlier, it is less commonly employed in practical applications. In contrast, feature-based image matching algorithms are better suited to complex real-world environments, offer higher robustness and scalability, and have broader applicability, making them more widely used in practice [100].

The SIFT operator is a classic algorithm in the field of feature extraction and image matching, first proposed by David Lowe in 1999 [101]. In 2004, the algorithm was further refined and enhanced [102]. Mikolajczyk et al. conducted experimental analyses on representative feature extraction operators, such as the SIFT operator and the Harris operator, considering factors like scale variation, and concluded that SIFT achieved the best extraction performance. However, SIFT suffers from high computational complexity, prompting numerous researchers to propose improvements, including SSIFT [103], PCA-SIFT [104], and CSIFT [105], among others. Herbert Bay and colleagues improved upon the SIFT algorithm and proposed the SURF algorithm, which significantly increased processing speed [106]. P. M. Panchal et al. compared the SURF and SIFT algorithms, finding that SURF achieved a performance comparable to SIFT [107]. The Oriented FAST and Rotated BRIEF (ORB) algorithm, proposed as an effective alternative to SIFT or SURF, demonstrated experimental performance approximately two orders of magnitude faster than SIFT [108], though with slightly lower robustness, sensitivity to illumination changes, and relatively reduced accuracy. Li et al. [109] optimized the threshold calculation of the ORB algorithm using grayscale statistical characteristics, effectively improving the stability of feature matching. Dai et al. [110] proposed an improved strategy that combines image enhancement and adaptive threshold fusion; the experimental results showed that this method increased the number of feature points extracted by about 30% under low-light and high-contrast conditions, and its matching accuracy and real-time performance both surpassed those of the traditional ORB algorithm. To guide optimal algorithm selection for different landslide environments,

Table 3. Comparison of feature matching algorithms for different landslide environments.

Algorithm	Accuracy	Speed	Memory Usage	Best Applicability	Limitations
SIFT	High	Medium	High	Vegetation-covered areas; complex textures	High computational cost
SURF	High	Fast	Medium	Exposed slopes; moderate vegetation	Less robust to extreme lighting changes
ORB	Medium	Very fast	Low	Real-time monitoring; bare ground	Lower accuracy in complex scenes

provides a comprehensive comparison of feature matching algorithms commonly used in UAV-based landslide monitoring. This table shows that SIFT achieves higher matching accuracy in vegetation-covered terrains due to its scale-invariant properties and robust feature detection, making it suitable for complex slopes with dense vegetation. By contrast, ORB offers substantial speed advantages for real-time monitoring, which is particularly beneficial for emergency response scenarios where rapid assessment is crucial. For exposed slopes with minimal vegetation, SURF provides an optimal balance between accuracy and computational efficiency.

Although SIFT, SURF, and ORB algorithms each have their shortcomings in certain aspects, they are interrelated and occupy significant positions in the field of image matching. Each has its own advantages, providing a variety of options for feature extraction and image matching.

SfM reconstructs camera extrinsic parameters such as position and orientation, as well as camera calibration data including focal length and radial distortion, by finding correspondences between images, thereby recovering 3D information from two-dimensional image pairs [111]. MVS reconstructs dense 3D geometric structures by searching for visual correspondences in images using the estimated camera parameters; these correspondences are triangulated to generate 3D information [112]. When integrated, SfM and MVS can fully automate the production of high-resolution Digital Elevation Models (DEMs) [113].

By applying SfM to multiview images with sufficient overlap, high-precision 3D models can be generated [111]. The SfM algorithm first uses multiview images with adequate overlap [114] and, based on image feature recognition techniques such as SIFT, can automatically detect, describe, and match homologous points between stereo pairs. According to the matched points and the camera’s intrinsic and extrinsic parameters, algorithms such as bundle adjustment are applied to estimate the 3D coordinates of each feature point, resulting in a sparse 3D point cloud. Unlike traditional photogrammetry, SfM can automatically reconstruct 3D meshes with known positions without requiring prior information [115]. Subsequently, methods such as MVS, deep learning, and semi-global matching are employed to densify the point cloud. Ground control points are then incorporated to georeference the 3D model within a real-world coordinate system. Finally, the reconstructed model can be exported as a DEM or, through projection, as a Digital Orthophoto Map (DOM). Compared with traditional aerial photogrammetry at equivalent resolution, the integration of SfM, MVS, and UAV platforms offers a cost-effective approach to generating high-resolution DEMs [116]. These products are widely applied in different fields: DEMs for Geographic Information Systems (GISs), DOMs for cartographic mapping, and DSMs for Building Information Modeling (BIM) [117].

Gupta et al. [112] utilized high-resolution landslide images acquired via UAVs, applied the SIFT algorithm for keypoint feature matching, and performed bundle block adjustment to precisely estimate the 3D position and orientation of the camera as well as the spatial coordinates of feature points, thereby constructing a sparse point cloud. MVS clustering was then used to densify the point cloud, followed by Poisson surface reconstruction to complete the 3D surface model. The open-source software CloudCompare was subsequently employed for 3D visualization and analysis. This study integrated UAV imagery with 3D reconstruction algorithms, providing a flexible and effective tool for landslide mapping and monitoring. Lucieer et al. [18] collected multiview high-resolution RGB images of landslides using UAVs equipped with standard digital cameras and GPS, generated 3D point clouds and DEMs using SfM and MVS techniques, validated SfM accuracy with ground control points, and applied the Co-Registration of Optically Sensed Images and Correlation (COSI-Corr) algorithm to analyze landslide displacement. Their results demonstrated that UAV imagery combined with SfM–MVS techniques enables accurate 3D surface reconstruction and provides a reliable basis for flexible landslide monitoring.

3.2. Three-Dimensional Reconstruction Based on UAV LiDAR

LiDAR is a widely used and important technology for acquiring 3D point clouds. Compared with UAV image-based modeling, airborne LiDAR enables more straightforward data acquisition and generates point clouds with higher accuracy, allowing for data collection and processing without relying on ground control points [118]. The accuracy of the 3D coordinates in point cloud outputs generated by airborne LiDAR systems primarily depends on the positional information provided by inertial sensors and satellite positioning sensors while also being influenced by system performance and external environmental conditions. The point cloud data, consisting of a large number of 3D coordinate points, can capture the shape and spatial distribution of target objects. Through processing steps such as denoising, filtering, and registration, and employing algorithms including Poisson reconstruction and Delaunay triangulation, these data can be converted into triangular mesh models or other 3D model forms, thereby facilitating the 3D reconstruction of target objects [119].

For landslide studies, LiDAR-based 3D scene models are measurable and enable the direct generation of contour lines, calculation of earthwork volumes, and support for disaster management and risk assessment. UAV-mounted LiDAR provides high-resolution terrain data and can penetrate dense forest canopies to detect landslide deposits, thereby offering enhanced precision and accuracy for large-scale landslide mapping and modeling [120]. However, its limitations include high equipment and maintenance costs, as well as the complexity and time demands of data processing, often necessitating its combined use with other methods. Nevertheless, LiDAR remains one of the most powerful and indispensable tools in contemporary landslide research.

Table 4. Comparison of key parameters between LiDAR and oblique photogrammetry.

Indicators	LiDAR	Oblique Photogrammetry
Accuracy	High (usually 2–5 cm, affected by point density, vegetation, etc.)	Low (usually 5–10 cm, affected by image resolution, overlap, control points, etc.)
Vegetation Penetration Capability	Strong	Weak
Data Volume	Large (requires noise reduction)	Small (dependent on image quality)
Cost	High	Low

presents a comparative summary of the critical parameters between LiDAR and oblique photogrammetry technologies.

3.3. Three-Dimensional Reconstruction Based on Hybrid Methods

In the era of big data, multi-source data fusion has emerged as a prominent trend in research and application. To better exploit the strengths of different data acquisition methods, overcome the inherent limitations of single approaches, and enhance reconstruction accuracy and quality for diverse and complex application scenarios, hybrid 3D reconstruction has emerged. Hybrid 3D reconstruction integrates multiple types of data and reconstruction techniques to generate comprehensive and high-fidelity 3D models.

Airborne LiDAR and UAV photogrammetry are currently the two primary methods for acquiring 3D spatial data in low-altitude remote sensing. However, each method has its own advantages and disadvantages in landslide modeling [121]. UAV imagery provides detailed texture and spectral information, but factors such as lighting variations and occlusions and image-based modeling may suffer from geometric distortions, reduced accuracy, and inconsistencies with the real scene [122]. LiDAR point clouds, in contrast, can accurately provide the 3D spatial position of objects [121], but LiDAR-based 3D reconstruction may lack sufficient resolution in regions with rich fine-scale details, potentially leading to the loss of subtle features. By fusing LiDAR data with UAV imagery, it is possible to obtain DEMs with higher coordinate accuracy [122], achieve the complete and unified recovery of weakly textured surfaces, build high-precision 3D models, and enable more accurate landslide monitoring, thereby providing critical scientific and technological support for research on landslide mechanisms, stability assessment, and emergency response [36].

The fusion of LiDAR and optical imagery generally utilizes algorithms such as Iterative Closest Point (ICP) to align LiDAR point clouds with images, achieving registration errors typically within 5 cm. During fusion modeling, weighted fusion strategies are commonly applied—for example, assigning 60% weight to point cloud geometry and 40% to image texture—to balance structural completeness with visual fidelity. Figure 6 illustrates a typical workflow for LiDAR–image fusion-based 3D reconstruction. Previous studies [123] report that compared to single methods, fusion approaches can improve accuracy by 20–30% in low-vegetation environments and by 30–40% in bare-ground conditions.

In one study [122], LiDAR point clouds and UAV imagery were combined: the LiDAR point cloud was first registered with the DSM generated from dense UAV image matching; then, high-quality registration points were selected as control points and incorporated into the UAV aerial triangulation solution. Through iterative refinement, a high-precision DSM was obtained. Furthermore, to reduce the influence of low vegetation on landslide terrain analysis, a DSM filtering algorithm based on irregular triangulated networks and slope was proposed to derive a high-accuracy DEM. DEM differencing was subsequently applied to analyze changes in landslide topography. Another study [36] integrated UAV imagery and LiDAR point clouds by combining hole-filling techniques based on partial differential equations (PDEs) with the non-uniform rational B-spline (NURBS) method to accelerate 3D scene reconstruction. Compared with least squares plane fitting, Delaunay triangulation, and polygon mesh methods, realistic 3D landslide models were generated through texture mapping. However, since LiDAR point clouds do not provide texture information, constructing refined models that meet both geometric and visual requirements still requires the acquisition of high-resolution ground photographs or the use of model repair software [119]. In another case study [124], 3D modeling of the study area was initially performed using UAV oblique photogrammetry and then supplemented with terrestrial laser scanning (TLS) data, resulting in a more complete 3D model.

Constructing refined and high-precision 3D landslide models is crucial forinvestigating landslide mechanisms, monitoring, early warning, and disaster mitigation. Accurate 3D models obtained through data fusion provide clear textures, rich details, and quantitative measurements. Although integrating point cloud and UAV oblique photogrammetry technologies offers comprehensive modeling capabilities, the large data volume results in longer processing times and higher costs, making it more suitable for the high-precision reconstruction of small-scale areas. While hybrid 3D reconstruction is less practical for large-scale landslides with high timeliness requirements and its application remains relatively limited, the fact that most landslide disasters in China are small-scale indicates that this approach still holds significant potential for future development.

4. Applications of UAV 3D Reconstruction in Landslides

4.1. Application of UAV-Based 3D Reconstruction in Landslide Emergency Investigation

Emergency investigation is a fundamental component in the management of landslide disasters and plays a critical role in both prevention and post-event assessment. It refers to the rapid acquisition of essential information, evaluation of hazard conditions, and formulation of emergency countermeasures following sudden events, thereby providing a robust scientific basis for disaster monitoring, mitigation, and assessment [125]. In the context of landslide emergency response, it is imperative to rapidly acquire data on the geological structure and environmental conditions of the affected area, emphasizing that emergency operations must be both “rapid” and “efficient” [126]. Recent advances in technology have significantly enhanced this efficiency. For instance, Chen et al. [127], targeting the critical “72-h rescue window”, proposed a UAV-based real-time processing pipeline: onboard Convolutional Neural Networks (CNNs) directly extracted landslide features, and through Residual Network (ResNet) feature fusion combined with 8-bit quantization compression, the inference time for 6000 × 4000 images was reduced from 109.47 s to 4.75 s—faster than the 5.13 s camera capture interval. This breakthrough enabled real-time computation during flight and immediate decision-making during transmission, thereby meeting urgent demands for low-latency, lightweight, and end-to-end UAV systems at disaster sites. Likewise, Zhu et al. [128] introduced an integrated “UAV–cloud” real-time identification system designed for shallow, small-scale landslides triggered by heavy rainfall in South China. Following the UAV acquisition of centimeter-level imagery, U-Net-based edge-enhanced segmentation and boundary extraction were completed within 30 s/km². The accompanying “UAV Cruise Geological Hazard AI Identification System” software then automatically delivered high-precision landslide boundaries (mean Intersection over Union (mIoU) of 90.7%), enabling near real-time, end-to-end UAV-based decision support for emergency evacuation and engineering rescue.

In recent years, UAV oblique photogrammetry has emerged as an indispensable tool for landslide emergency investigations due to its ability to rapidly, efficiently, and safely acquire high-resolution spatial data even in hazardous terrains [54,129]. For example, Kyriou, Nikolakopoulos, Kouk-ouvelas, and Lampropoulou [130] employed UAVs to acquire multitemporal datasets and, in combination with GNSS measurements, systematically evaluated landslide activity over time. By constructing high-fidelity 3D models of the affected areas, they directly quantified key parameters, including landslide extent and earthwork volume, thereby enabling the precise delineation of deformation boundaries [97].

Effectively identifying potential landslide hazards in advance and monitoring their spatiotemporal evolution to prevent catastrophic events remains a significant challenge in current prevention and control efforts [126]. To address this issue, UAV 3D reconstruction technologies have been increasingly adopted. For instance, Zhao et al. [125] conducted an emergency response study of the revived Gu’an Niangzhai landslide in Danba County, Sichuan, utilizing multirotor UAVs combined with total stations and near-real-time 3D modeling workflows. Through multi-source data fusion, the landslide body was subdivided into four deformation sub-zones, and 20 days of continuous monitoring confirmed that the slope had entered a uniform deformation stage. This research highlighted the necessity of integrating long-term stability monitoring with early warning mechanisms, providing a replicable framework for understanding landslide reactivation under complex geological conditions. Similarly, Zhao et al. [125] carried out an emergency investigation of the Liping Beishan landslide–debris flow in Zhouqu County, Gansu Province, using small UAVs. By rapidly acquiring high-precision orthophotos, this study accurately characterized landslide morphology, debris flow channels, and deformation features. Compared with traditional field surveys, this UAV-based approach improved efficiency by more than 60% while significantly reducing safety risks, highlighting the comprehensive advantages of UAV technology in landslide emergency response, particularly in terms of efficiency, safety, and precision.

To gain deeper insight into landslide dynamics, Zhao, Han, Liu, Zuo, and Zheng [37] constructed a high-resolution 3D landslide model using close-range photogrammetry integrated with detailed ground surveys to investigate landslide formation mechanisms. This approach not only enabled the quantitative extraction of morphological and deformation parameters but also provided comprehensive, multi-dimensional datasets to inform prevention strategies and engineering design. It further highlighted the advantages of UAV aerial surveys, characterized by “short acquisition cycles and high operational flexibility”, demonstrating significant value for detailed investigations at sensitive sites such as booster stations. Likewise, Li, Sun, Duan, Wang, et al. [131] optimized image acquisition angles and overlap rates to generate high-precision 3D models of landslides in reservoir areas, confirming the effectiveness of UAV oblique photogrammetry as a practical tool for emergency investigations. In addition, Galve et al. [132] applied UAV photogrammetry to the A-7 highway landslide in Spain. By sequentially constructing DEMs and 3D reconstructions, they quantitatively estimated the landslide volume and accurately projected the highway reopening time, thereby delivering reliable technical support for both immediate-response operations and post-disaster recovery planning.

Collectively, these studies highlight the broad applicability of UAV 3D reconstruction in landslide emergency investigations, demonstrating its ability to strengthen government-led disaster prevention and mitigation efforts while providing a rigorous scientific foundation and practical reference framework for future emergency management practices.

4.2. Application of UAV-Based 3D Reconstruction in Landslide Monitoring

Zhao & Lu [133] emphasized that landslide monitoring technology is essential for understanding the dynamic processes of slope movement, identifying early signs of deformation, and mitigating landslide-related risks. Effective monitoring and early warning systems are pivotal for minimizing casualties and economic losses [37]. In landslides at the pre-failure stage, the choice of emergency monitoring methods can critically determine the effectiveness of response measures [134]. Traditional manual techniques, such as precise geodetic surveys, can provide valuable data but are constrained by low efficiency, limited spatial coverage, and significant safety hazards. In recent years, UAV-based technologies have advanced rapidly, particularly with the maturation of Structure-from-Motion (SfM) algorithms, offering innovative, flexible solutions for real-time landslide monitoring. These approaches are especially advantageous in complex terrain, where UAV 3D reconstruction demonstrates considerable potential and broad applicability [18].

Compared with traditional 3D GIS approaches, UAV 3D reconstruction technology enables the generation of high-precision DEMs and DOMs, faithfully capturing micro-topographic features, accurately extracting attribute information, and estimating landslide volumes, representing a significant breakthrough for landslide monitoring and early warning [24].

Table 5. Technical comparison between UAV and traditional methods.

Indicators	UAV Monitoring	Traditional GNSS Monitoring
Monitoring point density/coverage	Continuous areal coverage	Discrete point monitoring
Cost (10,000 CNY/km2)	0.5–1	2.5–5
Spatial Resolution (UAV)/Point Spacing (GNSS)	0.1–0.3 m (depends on image resolution)	Usually 5–50 m (depends on deployment strategy and cost)
Data acquisition time	Fast (hourly, depending on area size)	Installed points: real-time; new points: hourly to daily (depends on terrain, accessibility)
Data processing and preliminary analysis time	Variable (hourly to daily, depending on data volume, method)	Installed points: near real-time; new points: included in deployment
Overall emergency response time	Usually several hours to 2 days (depends on task complexity)	Installed points: minutes to hours; new points: several hours to several days
Application scenarios	Large-scale areas, complex terrain, dangerous areas	Single point monitoring
Penetration capability	Weak (lower than LiDAR sensors)	None

presents a technical comparison between UAV-based and conventional monitoring methods. The ability to observe mountain landslides with high spatial and temporal resolution imagery, and to rapidly delineate affected areas, is essential for effective emergency disaster management. Given that landslides occur across multiple spatial and temporal scales [63], determining the minimum detectable area is critical. Small-scale landslides, which can easily evolve into larger hazards under heavy rainfall, necessitate precise monitoring and rapid identification to enable timely mitigation and informed decision-making [17]. In this context, UAV 3D reconstruction demonstrates unique strengths. For instance, Xiong [82] applied UAV oblique photogrammetry and 3D modeling to standardize the preprocessing of landslide monitoring data and establish risk assessment levels. The results showed an overall monitoring accuracy exceeding 99% and stability coefficients above three at all test sites, providing a robust scientific basis for improving both monitoring reliability and system resilience.

The surface morphology and overall deformation of landslides are critical parameters for effective landslide monitoring [69]. In this regard, numerous studies have proposed diverse methods and approaches. For example, Delbridge, Bürgmann, Fielding, Hensley, and Schulz [135] employed Unmanned Aerial Vehicle Synthetic Aperture Radar (UAVSAR) to characterize 3D surface deformation across the spatial extent of landslides, thereby enhancing both the accuracy and efficiency of deformation monitoring through multi-source data integration. Xie et al. [136] proposed a UAV flight path planning algorithm for landslide monitoring based on the Unreal Engine simulation framework, validating the fidelity of simulated trajectories and the robustness of the proposed algorithm. The experimental results indicated that the new algorithm achieved a 3D model simulation accuracy of 10–14 cm, with Root Mean Square Error (RMSE) values of 10–11 cm (representing a 2–3 cm improvement over conventional approaches) and a 9.3% reduction in flight time, thus offering more efficient and reliable technical support for landslide hazard monitoring. Pardeshi et al. [24] developed a method for extracting the 3D displacement field of landslides based on UAV photogrammetry, utilizing images, dense point clouds, and DSMs to derive displacement information. The image- and point-based method achieved an accuracy of 0.1 m, outperforming DSM-based approaches. These results demonstrate that densely matched 3D points from UAVs can serve as a viable alternative to LiDAR point clouds for determining displacement fields and angular variations, thereby providing more comprehensive information for landslide monitoring. Additionally, Zhao et al. [37] integrated close-range photogrammetry, big data analytics, internet technologies, and artificial intelligence to construct intelligent monitoring and early warning systems for landslides in mountainous areas of Guizhou, significantly enhancing both monitoring efficiency and predictive accuracy. Collectively, these studies not only provide innovative avenues for advancing landslide monitoring technologies but also furnish robust technical support for disaster prevention, risk assessment, and emergency response.

To better delineate the applicability of different UAV–sensor configurations in landslide monitoring, this study synthesizes and ritically compares three recent investigations conducted on the same landslide site. Gómez-Gutiérrez & Gonçalves [137] deployed both fixed-wing UAVs equipped with LiDAR and multirotor UAVs with optical sensors in a Portuguese coastal landslide area. The fixed-wing+LiDAR system achieved single-flight coverage of 1.2 km² with a volume estimation error of only 4.8%, whereas the multirotor+visible light system covered just 0.2 km², exhibiting an increased error of 12.1% but a 60% reduction in operation time. Building on this, Liao et al. [138] quantified performance in a Guizhou landslide site, reporting LiDAR vertical RMSE values of 3–5 cm compared to 7–9 cm for multirotor visible light SfM–MVS. Sun et al. [27] further reviewed that fixed-wing UAVs, with their longer endurance, are more suitable for large-scale reconnaissance, while multirotor UAVs, despite shorter endurance, excel at fine-scale crack detection on high-risk slopes and offer relatively lower costs. Collectively, these findings suggest that fixed-wing+LiDAR systems should be prioritized for large-area, high-accuracy surveys, whereas multirotor+visible light/SfM–MVS platforms are better suited for small-scale, high-precision crack monitoring or rapid emergency response.

4.3. Application of UAV-Based 3D Reconstruction in Disaster Assessment

Landslide disaster assessment is a critical component in disaster response, focusing on the rigorous scientific analysis of landslide magnitude, damage extent, triggering factors, and potential secondary hazards. The ultimate goal is to establish a robust basis for subsequent disaster management and post-disaster reconstruction. In this context, UAV 3D reconstruction technology, renowned for its distinctive advantages in rapid, high-precision, and intuitive spatial data acquisition, demonstrates a significant application value in disaster assessment. By generating detailed 3D models, it becomes possible to precisely and comprehensively identify the key morphological features of landslides, including gullies, collapses, and surface cracks. These features provide indispensable evidence for understanding landslide formation mechanisms and evaluating potential hazard risks.

Unlike conventional approaches that often fail to deliver timely 3D information, UAV 3D reconstruction effectively addresses this limitation and has increasingly emerged as a core technique in landslide disaster assessment [24]. For instance, Teo, Fu, Li, Weng, & Yang [55] applied UAV low-altitude photogrammetry to generate DOM, DTM, DEM, and 3D models of the affected area. By calculating volumetric differences between the source and deposition zones, they estimated the total landslide volume, thereby delineating the precise spatial extent and quantifying its hazard potential. Such analyses are crucial for supporting targeted emergency responses and for minimizing casualties and property losses caused by geological hazards.

Additionally, Ahmad, Shafique, & Hussain [139] analyzed the spatial distribution of landslide volume changes using UAV imagery and derived digital surface models (DSMs) in conjunction with the COSI-Corr algorithm, thereby providing a rigorous scientific basis for identifying high-risk landslides and informing policymakers in the development of effective mitigation strategies. Furthermore, Zheng et al. [140] integrated ground-based interferometric synthetic aperture radar (GB-InSAR), TLS, and UAV photogrammetry to conduct 3D modeling of major landslide bodies. Within just 48 h, they mapped the landslide expansion zone and completed a comprehensive hazard assessment, offering practical solutions for landslide emergency response and disaster evaluation.

Similarly, Vivaldi et al. [23] employed UAV oblique photogrammetry to capture high-resolution imagery and, combined with ground control points and UAV Position and Orientation System (POS) data, performed fully automated 3D reconstruction of the landslide. This process generated high-precision products including DEM, DOM, and DSM, enabling slope, aspect, and profile analysis, and integrated geotechnical strength indices to comprehensively evaluate landslide scale and hazard potential. The study further established a standardized post-disaster database to facilitate the efficient management and utilization of landslide information. Liang et al. [141] produced a high-resolution (5 cm) 3D model of the landslide area using UAV oblique photogrammetry and accurately calculated the landslide volume at 4.5 million cubic meters with an error margin of only ±3%. These precise datasets proved essential for post-disaster insurance claims and reconstruction fund allocation, underscoring the scientific reliability and practical value of UAV-based rapid landslide investigation and assessment.

In summary, the application of UAV 3D reconstruction technology in landslide disaster assessment has not only significantly enhanced both the accuracy and efficiency of assessments but also robust technical support for emergency response and disaster risk management. With the continuous development and refinement of related technologies, it is expected that UAV 3D reconstruction will play an increasingly critical role in future landslide disaster assessments.

Nevertheless, the widespread application of UAV 3D reconstruction in landslide monitoring remains limited by regulatory permits, certifications, and operational requirements. Globally, UAV regulations differ substantially across regions, with authorities adapting frameworks to their specific contexts and operational needs. In the United States, for example, operations are governed by the Federal Aviation Administration (FAA) Part 107 Small Unmanned Aircraft System rules: operators must obtain Part 107 Remote Pilot certificates, maintain visual line of sight throughout operations, avoid flights over people or moving vehicles, and ensure on-site visibility of no less than 3 statute miles. When beyond visual line of sight (BVLOS), missions are required for large-scale landslide monitoring, and operators must apply for exemptions—these can take weeks to months and may substantially delay emergency responses. In Europe, the European Union Aviation Safety Agency (EASA) requires operators to maintain visual line of sight and complete a Specific Operations Risk Assessment (SORA) in advance; in densely populated or sensitive areas such as the Alps, BVLOS approvals involve additional procedures, often requiring third-party safety demonstration agencies, further prolonging the approval process. In China, the Civil Aviation Administration of China (CAAC) categorizes mapping-grade UAVs as belonging to the “specific category”, requiring real-name registration, pilot licenses such as the Unmanned Aerial Vehicle Training Certificate (UTC) or Aeronautical Sports Federation of China (ASFC) certification, and airspace applications at least one day in advance. Collectively, these regulatory constraints can can compromise the inherent timeliness advantages of UAVs for rapid landslide emergency response.

5. Challenges and Future Perspectives

Although UAV 3D reconstruction technology demonstrates broad application potential in landslide emergency investigations, monitoring, and disaster assessment, it still faces complex technical challenges, including the following: (i) Inherent limitations of UAV platforms: A major constraint is limited flight endurance, primarily restricted by current battery technology. Multirotor UAVs typically sustain only 20–30 min of flight per mission, and when carrying LiDAR payloads, endurance drops to about 20 min—representing a 40% reduction compared to unloaded flights. This significantly affects monitoring continuity and spatial completeness, particularly in large-scale missions, thereby limiting the ability to comprehensively capture landslide dynamics. Furthermore, restricted payload capacity limits the number and type of high-precision, multi-source sensors that can be simultaneously integrated. (ii) Massive data volume and heterogeneity: UAV-acquired datasets are both voluminous and heterogeneous, greatly increasing the difficulty of processing. For instance, processing 1 km² of landslide monitoring data requires 8–12 h, with point cloud denoising alone consuming up to 40% of total processing time. Such demands require efficient processing and timely analysis. In particular, the precise registration and effective fusion core technical bottlenecks of multi-source data (e.g., optical imagery and LiDAR point clouds) that urgently require innovative solutions. (iii) Data processing and analysis bottlenecks: The vast and complex nature of UAV-collected data makes processing and analysis challenging, particularly regarding the accurate matching and efficient fusion of multi-source data, which remains a pressing technical problem. (iv) Restricted coverage and operational complexity: A single UAV flight typically covers less than 30 km². Monitoring larger areas necessitates multiple missions and/or securing a BVLOS waiver, substantially complicating mission planning and execution. (v) Environmental and weather constraints: Adverse meteorological conditions often restrict or even prevent UAV operations, reducing reliability in emergency scenarios. (vi) Regulatory hurdles: Regulations vary significantly across countries, requiring operators to thoroughly understand local frameworks and obtain the necessary permits. This process can be complex and time-consuming, thereby hindering cross-border research collaboration and delaying rapid-response investigations.

To address these challenges, future research and development could focus on the following key directions: (i) AI- and machine learning (ML)-driven data processing: Developing more efficient algorithms and software to improve automation levels, while exploring the use of artificial intelligence (AI) and ML [16] for the automated extraction and identification of landslide features (for example, CNN algorithms have achieved an identification accuracy of 92.5% for natural terrain landslides [142]; the Long Short-Term Memory (LSTM) model predicts 30-day landslide displacement with an error of ≤5 cm, while the multivariable LSTM model achieves 6-year displacement predictions with mean square error, rmse, and mean absolute error of 0.64, 0.80, and 0.50 mm, respectively [143]), thereby reducing both computational costs and processing time. (ii) Standardization and multi-source data fusion: Establishing standardized methodologies for data integration and processing to enable the deep fusion of UAV data with other remote sensing (e.g., satellite) and ground-based measurements (e.g., GNSS, in situ monitoring stations). This requires optimizing algorithms and building unified data interaction frameworks, thereby improving the comprehensiveness, precision, and reliability of landslide monitoring and analysis. (iii) Regulatory evolution and operational flexibility: Anticipated regulatory updates (such as the proposed FAA Part 108 in the United States) are expected to provide greater flexibility for BVLOS operations, substantially enhancing the feasibility of UAVs for large-scale and rapid landslide mapping. (iv) Optimizing UAV platform performance: Advancing new high-performance battery materials to increase energy density and flight endurance (e.g., solid-state batteries can extend multirotor endurance to 45 min, thereby meeting continuous monitoring demands for most landslide applications and reducing the need for frequent return-to-base missions); optimizing UAV structural designs and power systems to increase payload capacity and efficiency while reducing weight; and integrating precise wind measurement and prediction systems with advanced flight control algorithms to improve operational stability; for example, carbon fiber airframes reduce weight by 30% while simultaneously enhancing wind resistance, lowering energy consumption in complex terrains, and ensuring stable imaging under strong airflow conditions. (v) Developing integrated monitoring systems: Employing a comprehensive monitoring framework that combines UAVs with ground-based stations and GNSS, thereby leveraging the complementary strengths of different technologies to establish a more robust and resilient landslide monitoring network.

6. Conclusions

This review provides a systematic examination of the application and development of UAV 3D reconstruction technology in landslide disaster research. By analyzing the characteristics of different UAV platforms and the capabilities and limitations of onboard sensors, comparing the principles, workflows, and application scenarios of oblique photogrammetry- and LiDAR-based reconstruction methods, and discussing the potential of hybrid approaches that leverage their complementary strengths, this review offers a comprehensive overview of the current technical landscape. It further details the specific applications and demonstrated effectiveness of this technology in three core areas, landslide emergency investigation, dynamic monitoring, and disaster assessment, emphasizing its unique role in acquiring high-precision spatial information and supporting rapid, evidence-based decision-making. Simultaneously, this review objectively addresses key challenges, including limited flight endurance, restricted spatial coverage, susceptibility to environmental disturbances, high computational demands for data processing, and regionally variable regulatory frameworks.

Looking ahead, several research and development priorities emerge: optimizing UAV platform performance, advancing AI-driven automated data processing and feature extraction, promoting standardized multi-source data fusion, and enhancing integrated air–ground monitoring networks (e.g., fixed-wing UAVs for large-scale scanning, multirotor UAVs for local detail acquisition, and ground stations for continuous observation). In addition, adapting to evolving BVLOS regulations will be crucial for enabling large-scale, real-time landslide mapping and monitoring. Overall, this review aims to serve as a systematic reference for researchers to understand the UAV-based 3D reconstruction framework in landslide studies, recognize current technical bottlenecks, and identify future research directions.