Next Article in Journal
Development of a Methodology Based on ALS Data and Diameter Distribution Simulation to Characterize Management Units at Tree Level
Previous Article in Journal
An Assessment of the Seasonal Uncertainty of Microwave L-Band Satellite Soil Moisture Products in Jiangsu Province, China
Previous Article in Special Issue
From Data to Decision: Interpretable Machine Learning for Predicting Flood Susceptibility in Gdańsk, Poland
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 16
Issue 22
10.3390/rs16224236
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

The DEM Registration Method Without Ground Control Points for Landslide Deformation Monitoring

by
Yunchuan Wang
,
Jia Li
*,
Ping Duan
,
Rui Wang
and
Xinrui Yu
Faculty of Geography, Yunnan Normal University, Kunming 650500, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2024, 16(22), 4236; https://doi.org/10.3390/rs16224236
Submission received: 12 October 2024 / Revised: 9 November 2024 / Accepted: 12 November 2024 / Published: 14 November 2024
(This article belongs to the Special Issue Advances in GIS and Remote Sensing Applications in Natural Hazards)
Browse Figures
Versions Notes

Abstract

:
Landslides are geological disasters that are harmful to both humans and society. Digital elevation model (DEM) time series data are usually used to monitor dynamic changes or surface damage. To solve the problem of landslide deformation monitoring without ground control points (GCPs), a multidimensional feature-based coregistration method (MFBR) was studied to achieve accurate registration of multitemporal DEMs without GCPs and obtain landslide deformation information. The method first derives the elevation information of the DEM into image pixel information, and the feature points are extracted on the basis of the image. The initial plane position registration of the DEM is implemented. Therefore, the expected maximum algorithm is applied to calculate the stable regions that have not changed between multitemporal DEMs and to perform accurate registrations. Finally, the shape variables are calculated by constructing a DEM differential model. The method was evaluated using simulated data and data from two real landslide cases, and the experimental results revealed that the registration accuracies of the three datasets were 0.963 m, 0.368 m, and 2.459 m, which are 92%, 50%, and 24% better than the 12.189 m, 0.745 m, and 3.258 m accuracies of the iterative closest-point algorithm, respectively. Compared with the GCP-based method, the MFBR method can achieve 70% deformation acquisition capability, which indicates that the MFBR method has better applicability in the field of landslide monitoring. This study provides an idea for landslide deformation monitoring without GCPs and is helpful for further understanding the state and behavior of landslides.

Graphical Abstract

1. Introduction

Landslides are geological disasters with great destructive power that greatly affect the safety of people, as well as social, industrial, and agricultural production [1,2,3,4]. In landslide research, landslide surface deformation is becoming a key breakthrough point. The deformation of landslides can provide relevant information and parameters such as morphological characteristics, dynamic evolution, and early identification, which are conducive to the prevention and treatment of landslides [5,6,7,8,9]. Therefore, the effective acquisition of landslide deformation information is important for landslide investigation and prevention.
There has been much research on landslide morphology detection and surface deformation, utilizing methods such as terrain curvature and indices such as the terrain wetness index (TWI) and terrain roughness index (TRI) [10,11]. These methods are often used to identify landslide areas and detect landslide morphology. Additionally, satellites such as Landsat, SPOT, and Sentinel, along with their related technologies, are employed to monitor surface deformation [12,13,14]. In recent years, new remote sensing technologies such as airborne LiDAR, close-range photogrammetry, and unmanned aerial vehicle (UAV) photogrammetry systems have significantly improved the detail and quality of terrain information, enabling the acquisition of digital elevation models (DEMs) with higher resolution and quality [15,16,17,18]. A digital elevation model, which contains many topographic elements, surface characteristics, and other information, is used for the quantitative monitoring and analysis of landslide surface morphological changes in some methods [19,20,21,22,23]. Among these methods, constructing a differential model is the most intuitive way to perform deformation monitoring, which requires multiple DEMs to be referenced in the same coordinate system [19,24,25,26]. Multitemporal DEMs may originate from different data sources and at different times, and common ground control points (GCPs) are often used for coordinate transformation into the same coordinate system. However, landslides are often sudden and targeted. It is impossible to arrange and measure GCPs in all areas before landslide disasters occur. Only after a landslide occurs can the landslide be evaluated and monitored by UAVs and other equipment. Additionally, the deployment of ground control points is a laborious task, and they are prone to displacement or damage with the sliding and collapse of the landslide. Thus, it is possible to face a situation where no control points are available. Moreover, the precise registration of DEMs without GCPs is very important for landslide deformation monitoring.
DEM registration without control points involves the transformation of different DEMs into the same coordinate system to solve the transformation parameters [27,28,29,30]. There are two main ideas for DEM registration, one of which is based on common control surfaces, that is, finding surfaces with invariant positions and features in different DEMs, on which spatial registration is performed [31,32]. The control surface selected by this method is often unique, so it is applicable only to the data of control surfaces with specific characteristics. It easily fails in areas where the characteristics are not obvious or lack unambiguity. Another idea involves DEMs in the registration process via the least squares technique, and registration parameters such as rotation and translation are solved through iterative calculations [33,34,35]. The most representative method is the iterative closest point (ICP), which takes the nearest point in space as the corresponding point and achieves registration by closing the spatial distance. The ICP algorithm is simple in principle and powerful in performance, balancing applicability and convenience, and remains one of the most mainstream choices today [10,36]. However, it has several constraining conditions in practice. First, it requires that the two surfaces have good initial alignment. Further, good registration accuracy can be achieved only when the points on the two surfaces are guaranteed to be in one-to-one correspondence. This makes the ICP algorithm problematic in practical applications, especially in the field of landslide deformation monitoring. The presence of differences in the monitoring data has an impact on the applicability of the alignment algorithm, and the difficulty in establishing the correct point-to-point relationship leads to low accuracy and poor practical monitoring results.
Therefore, the research objective of this paper is to effectively solve the multitemporal DEM registration problem without GCPs. A simple, fast, and effective landslide deformation monitoring solution that automatically identifies and selects stabilized areas can help to better understand and evaluate landslide conditions. The proposed method is called multidimensional feature-based coregistration (MFBR).

2. Methods

The workflow of MFBR consists of three stages, as shown in Figure 1: DEM construction, DEM registration, and landslide deformation monitoring. The first stage relies mainly on photogrammetry and computer-vision-related methods to generate a DEM from image observation data, which serves as the primary representation of the landslide area. The DEM registration stage can be divided into two steps. The initial position calculation, which transforms the elevation information of the DEM into image information, is based on the corresponding feature points to achieve coarse registration of the DEM data. The second step is to analyze the characteristics of elevation changes in landslide monitoring data during different periods and to pre-extract stable and deformed regions to implement precise registration based on stable regions. Finally, the landslide deformation information is obtained by constructing a differential model.

2.1. Multidimensional Feature-Based Coregistration Method

2.1.1. Two-Dimensional Image-Based DEM Position Registration

The principle of the two-dimensional registration method is shown in Figure 2 below, and its detailed process is as follows:
Stage a: The surface elevation features of DEMs can be transformed into images so that the image features can be extracted and the corresponding characteristic points can be identified. The advantage is that the attribute categories of feature points are not considered to prevent tedious discriminative and computational processes in the extraction of terrain feature points. Therefore, the DEM is first exported as a two-dimensional image, and the resolution of the image is consistent with that of the DEM.
Stage b: Since the UAV images can capture areas other than the deformation area, the extraction method of the corresponding characteristic points has better applicability here. The main goal of this section is to obtain the initial transformation relationship between different DEMs, which does not require extreme precision. Therefore, the faster speeded-up robust features (SURF) algorithm is used to extract feature points and reject incorrect correspondences by applying the random sampling consistency algorithm.
Stage c: The acquired corresponding characteristic points are transformed by the image coordinate system to the projection coordinate system, and the coordinate transformation parameters (the rotation matrix R and the translation matrix T) of the DEMs in different coordinate systems are solved. Finally, the DEMs to be registered are transformed by applying Formula (1) to achieve the initial registration between the data.
X Y Z = m × R × x y z + T
where m, R, and T are scale factors, rotation matrices, and translation matrices, respectively. Since the spatial resolutions of the DEMs in different periods are the same here, m is 1, [x y z] T is the original coordinate system coordinate, and [X Y Z] T is the transformed DEM coordinate.

2.1.2. Accurate Registration of DEMs on the Basis of 3D Spatial Features

The closest point in space is the corresponding point in the ICP algorithm; however, the spatial point pair relationship between two surfaces becomes complicated because of deformation in the measured data, and there may be more false correspondence points if the closest point in space is used directly as the corresponding point. To solve this problem, the expectation maximization (EM) algorithm [37,38,39] is introduced in the precision registration process to identify and reject the coarse deviations of DEMs before precision registration to obtain accurate precision registration results. A diagram of precise registration, which consists of two main steps, is shown in Figure 3. The detailed process is as follows:
Stage 1: Gross error identification and rejection based on the EM algorithm. When surface deformation occurs in a region, the elevation value corresponding to the DEMs is subtracted to obtain an obvious elevation difference. At this time, the elevation difference histogram must contain multiple peaks. This includes two different types of difference values: random error and surface deformation information. Therefore, it is assumed that the height difference histogram between the DEMs of two periods consists of multiple Gaussian distributions. While containing the implied variables, the EM algorithm is used to estimate the probability density functions of different regions by iterative means with great likelihood, and then the deformation gross error regions can be identified and eliminated according to the probability density distribution. The detailed process is as follows.
(a)
Regarding DEM position registration, the elevation value of one phase of the DEM is interpolated at the same location as the other DEM. The difference calculation is subsequently performed to obtain the height difference histogram.
(b)
The multitemporal DEM height difference histogram is assumed to be mixed by multiple Gaussian models, and two are assumed here. The initial values of the Gaussian models are set according to the height difference histogram.
(c)
Based on the initial model parameters or the model parameters calculated in the previous iteration, the current posterior probabilities are calculated as current estimates of the hidden variables.
(d)
Maximum likelihood estimation is performed according to the current value, and the likelihood function is maximized to obtain a new parameter value.
(e)
The iterative calculation is repeated until the desired model parameters converge. According to the distribution criterion in the Gaussian model, the region of [μ1 − 3σ1, μ1 + 3σ1] in Gaussian Model N1 is considered the point of the stable region.
Stage 2: Second, on the basis of the gross error elimination of the EM algorithm, the ICP algorithm [33,40] is used for precision registration. Through repeated iterative computations, this process aims to achieve the optimal spatial alignment of DEMs from different periods, minimizing their offset.

2.2. Landslide Deformation Monitoring Based on DEM Registration

The surface deformation for a certain period is obtained via the multiperiod DEM of difference (DoD) method. Since the DEMs generate certain errors during data processing, when performing the DEM differential operation, the differential values contain systematic errors and cannot be used directly for the calculation of form variables. Therefore, the minimum detection threshold (minLoD) [41,42] needs to be determined first to distinguish between the system noise and the actual surface variation. Since the MFBR method does not have ground control points and check points, the RMSE of the distances of corresponding grid points within the stability interval extracted by the MFBR method in the DEM of two periods is used as minLoD, and the expression is
min LoD = 1 n ∑ i = 1 n z i D E M a − z i D E M b 2
where DEMa and DEMb are the DEMs of different periods, zi is the corresponding elevation value, and n is the number of grid points.
On this basis, the deformation information is obtained through the difference model; that is, the corresponding elevation values between the multiperiod DEM data are subtracted, as shown in Formula (3).
D e = D E M a i , j − D E M b i , j

3. Study Area and Data

3.1. Experimental Data

Three groups of experimental data are set to verify the MFBR method: (1) To simplify the judgment and analysis of the experimental results, simulation datasets are created for testing, where the test conditions can be close to those of the actual application and can provide an objective evaluation. (2–3) Real landslide cases are used to obtain data from two landslides that occurred in Luchun County and Gongshan County. Landslide monitoring data in Luchun County were obtained after the occurrence of the landslides for post-disaster monitoring work, and landslide data in Gongshan County were obtained before and after the landslides occurred.
(1)
Simulation data
The simulation dataset was taken from the DEM of Yunnan Province, with a spatial resolution of 15 m and a size of 556 × 556. It has many typical features, such as mountains and rivers (as shown in Figure 4), and the maximum elevation difference is approximately 1000 m. The whole DEM is divided into nine uniform areas. To simulate the real deformation conditions in the simulated DEM data, the local noise N2 of (5–25) is added to the five areas of the simulated DEM data to simulate the deformation area. Moreover, the random Gaussian noise N1 with a mean of 0 and a variance of 1.8 is added to the remaining four areas to simulate the system error caused by the sensor and environment. The simulation DEM is subsequently rotated and translated according to the conditions of X = Y = Z = 15, a = b = c = 2°, and m = 1 to generate the simulation DEM data.
(2)
Landslide monitoring data from Luchun County
The landslide is located on the S315 highway beside Luchun County, with an area of approximately 0.049 km2. The length of the landslide reaches 493 m, and the height difference is approximately 170 m, as shown in Figure 5. Since July 2021, the site has been affected by continuous rainfall, resulting in a sudden landslide in August 2021, causing some damage to a wide range of roads and traffic disruptions. The landslide is located next to a road with a large slope angle and has a certain vegetation cover on the surface and a history of sliding phenomena. After the landslide, two periods of image data were acquired by UAVs in August and December 2021, and the DJI UAV and Pegasus UAV were used for aerial photography acquisition. The flight altitudes of the UAVs were set to 130 m and 110 m, and the average image resolutions were 3.41 cm/pixel and 5.71 cm/pixel for the two periods, respectively. For the ground survey, nine ground control points were deployed, and a Leica GS15 was used to set up a survey base station to obtain the WGS84 coordinates of the control points (EPSG: 4326). To test the feasibility of the method, the location information in the EXIF of the photos is not used here, so the two periods of data are in different spatial locations.
(3)
Landslide monitoring data from Gongshan County
In April 2022, a landslide disaster occurred in Dulongjiang Township, Gongshan County, Nujiang Lisu Autonomous Prefecture, Yunnan Province, China, which covered an area of approximately 0.68 km2, with the length of the landslide reaching 862 m and the difference in elevation reaching 600 m, as shown in Figure 6. The region is located in the Nujiang River deep-cutting alpine canyon river terrace area, with high mountains, steep slopes, loose soil, and other characteristics, and is highly susceptible to landslide disasters. The landslide left six people missing, buried two auto repair shops, a scrap iron recycling site, and two cars, washed out a hydrological station, and damaged some houses. After the landslide occurred, the Pegasus UAV was used to acquire image data of the landslide; the UAV flew at an altitude of approximately 230 m with an image ground resolution of 3.52 cm/pixel. The pre-landslide imagery was derived from a large-scale census in 2020 and was acquired using a Rainbow UAV flying at an altitude of approximately 3000 m, with an image ground resolution of approximately 21.6 cm/pixel. Owing to the large difference in image resolution between the two periods of data, to enable multiple DEMs to represent surface information with the same level of detail, the DEM data from both periods were resampled to the same resolution for better matching and deformation detection.

3.2. DEM Construction of UAV Images

The image was processed via Agisoft Metashape Pro 1.7 with an Intel 12th Gen i5-12600KF, NVIDIA RTX3070Ti, and 16 GB RAM. The structure-from-motion (SfM) calculation was run in Metashape pro 1.7, in which the internal parameters and relative orientation of the camera are estimated. The eponymous features in the images are identified and paired, and an image network is constructed to build a sparse point cloud. The geographic coordinates of the point cloud are provided with the location information recorded by the UAV at the time of filming, and then a dense point cloud of the study area is constructed using the MVS dense matching algorithm. The point cloud is filtered via the statistical outlier removal (SOR) algorithm to remove the outlier points generated in the previous step. Finally, inverse distance weighted interpolation (IDW) is used to interpolate the UAV image point cloud to construct a DEM for the study area. Here, the resolutions of the DEM data in Luchun County and Gongshan County were 0.25 m and 0.5 m, respectively.

3.3. Evaluation Metrics

The absolute registration accuracy of the simulated DEMs is verified by check points. A certain number of coordinate points (check points) are randomly selected in the nondeformed area of the simulated DEMs, and the elevation difference among the check points when manual error is added and no transformation is performed and the elevation difference after registration is completed are obtained as the absolute accuracy of the alignment of the simulated dataset. The root mean square error (RMSE) is used as a measure of the relative registration accuracy of the real DEMs and the simulated DEMs [43], and the formula for calculation is shown in Formula (4). The RMSE measures the spatial distance of all points between the DEM data that need to be matched in different periods, and the smaller the value of the RMSE is, the higher the alignment accuracy.
RMSE = 1 n ∑ i = 1 n x i − x i c o 2 + ∑ i = 1 n y i − y i c o 2 + ∑ i = 1 n z i − z i c o 2
where x, y, and z are the coordinate points in the reference DEM; xco, yco, and zco are the corresponding points of the above reference DEM points in the DEM to be matched; and n is the number of grid points.
To measure the accuracy of landslide deformation, the deformation detection capability is defined as ab(d):
a b d = S ∩ S ’ / S
where S and S′ are the real landslide deformation area and the detected landslide deformation area, respectively.

4. Results

4.1. DEM Registration Results

(1)
Simulated Registration Experiment
The registration results of the MFBR and ICP methods are compared according to the accuracy evaluation method in Section 3.3 above, as shown in Table 1. With more metrics, more objective accuracy results can be provided by the simulation data. With respect to the accuracy of the registration, the proposed method provides significant improvements over the classical ICP method, as the RMSE of the registration based on the MFBR is 0.963 m, whereas that based on the ICP is 12.189 m. The improvement in the MFBR registration scheme over the traditional ICP scheme is also reflected in the accurate determination of the check points. In the original data, owing to the addition of random error, the elevation value of points at the same plane position also has an error of 0.5 m. In comparison, the RMSE error in the MFBR registration result is 2.528 m, which improves the accuracy by approximately 2 m compared with the classical ICP method, confirming the feasibility and superiority of the MFBR.
(2)
Landslide registration in the real case
In this process, the positional information in the images is removed, placing the data from the two periods in independent local coordinate systems to verify the feasibility of the method. The multitemporal DEM data registration accuracies of landslides in Luchun County and Gongshan County are summarized in Table 2. For the Luchun County landslide data, the final registration accuracy after the MFBR method was used was 0.36 m, which was approximately 1.5 times greater than the DEM resolution, and the accuracy was greater for the deformation data. In response to the Gongshan County landslide, the overall accuracy is lower than that of the landslide in Luchun County. On the time scale, the time interval between the acquisition of landslide data in Gongshan County was approximately 2 years, more than 80% of the research area was covered with vegetation, and change in the growth of vegetation was relatively significant. At the spatial scale, the landslide data area in Gongshan County is more than 10 times larger than the landslide data area in Luchun County, and the elevation difference is larger and more drastically variable. At the data scale, there are large differences in image resolution, which is the main reason for the large differences in the representation of vegetation areas in the DEM; however, the MFBR method still improves the registration accuracy close to 1 m. The experimental results show that even though rough registration provides good initial alignment, the ICP algorithm always looks for the corresponding points on the basis of the nearest point in space, which means that the deformed grid points are involved in the process of establishing the corresponding points. As the deformation area increases, there are more corresponding point relationships that are wrong, which in turn makes the registration increasingly inaccurate.

4.2. Deformation Detection

After successful registration, deformation detection is carried out on actual data and simulation data. Figure 7, Figure 8 and Figure 9 show the results of the simulated data and the landslide data of Luchun County and Gongshan County, respectively, and the corresponding deformation detection capability values for the different experimental groups are shown in Table 3.
The deformation detection of the simulation data is shown in Figure 7, where a and b are the deformation detection results of the original untransformed data (which can be regarded as true values) and the registration using the MFBR method. The results based on the ICP method are not shown here because it fails to obtain the correct deformation information. As seen from the figure, owing to the random error added to the whole region, some smaller error (deformation) points cannot be found, resulting in the discrete distribution of deformation regions in the deformation results. In contrast, the extracted deformation region is very similar to the original region in Figure 7a after the MFBR registration method is used. In the accumulation deformation area shown in the blue area in the figure, there is only a 1 m difference from the true value, indicating that the result is relatively accurate. Combined with Table 3, the method has a deformation detection capability of 73.9% for these data, and this is under the condition that the original data are added to the random errors, which indicates that the MFBR method has a strong deformation detection capability.
The deformation detection of landslides in Luchun County is shown in Figure 8, where a, b, and c are the deformation detection results using ground control points and the MFBR and ICP methods. The GCP group has the most deformation areas monitored, followed by the MFBR method and the ICP method. Correspondingly, the deformation detection capabilities of MFBR and ICP are 83.54% and 60.69%, respectively. Comparing the three areas on the right side of Figure 8 show that the MFBR-based method is very close to the GCP results but fails to detect some smaller deformations, whereas the ICP-based method, in the same case, suffers from more areas of missing detections.
The landslide deformation detection results for Gongshan County are presented in Figure 9, with the corresponding detection performance summarized in Table 3. The MFBR method demonstrated a detection accuracy of 71.24% of the true deformation values. In the top region of this landslide, as shown in the upper right-hand-side comparison in Figure 9, both methods were able to capture deformation information in this region, but the MFBR-based method was able to capture more subtle changes. In the bottom region of the landslide, as shown in the middle of the lower right side of Figure 9, the MFBR method detected nearly all of the deformation, although a few small changes were missed compared with the true values. In contrast, the ICP method failed to detect most of the deformation in this area. These results highlight the effectiveness of the MFBR method in capturing detailed landslide deformation information, despite the reduction in registration accuracy due to factors such as temporal discrepancies, data inconsistencies, and complex vegetation cover.

5. Discussion

5.1. Stable Region Extraction Accuracy Analysis

For multiperiod DEM data with landslide deformation, the core of accurate registration is to locate the area accurately without deformation in the data to establish the correct multiperiod correspondence. Therefore, in the MFBR method, the accurate registration of multitemporal DEMs on the basis of 3D spatial features is different from that of other methods and is also the most critical step. The core idea is to distinguish and extract the deformation/roughness areas and the stable areas that can be used as the registration data on the basis of the histogram of the height difference distribution between DEMs. To verify the accuracy of the extracted area, a dense point cloud based on ground control points is constructed, and the stable area is viewed via the “cloud-to-cloud (C2C)” method using CloudCompare software 2.13. The results are shown in Figure 10. The left side of Figure 10 shows the landslide data from Luchun County, and the results of the two methods are in very strong agreement with each other. In these data, several major deformation intervals, such as part of the top area of the landslide and the upper right side, and stabilizing regions, such as the top road and the stepped surface on the right side, were accurately identified via the MFBR method. The quantification of stable regions shows that the MFBR method achieves a high extraction rate of 91.44%. On the right side of Figure 10, landslide data from Gongshan County are shown. The most significant deformation in this region occurs within the landslide body, along with some areas affected by vegetation. The MFBR method provides better extraction accuracy here as well, with a rate of approximately 73.20%. In the vegetation-covered area on the upper left side of the region and the middle–lower section, certain area has been lost. The landslide body in this region has undergone significant deformation, while the vegetation in the remaining areas has also experienced noticeable changes. This has led to a decrease in initial registration accuracy, and even small misalignments have resulted in errors when estimating elevation differences, thereby compromising the detection of some stable areas. In conclusion, the data from two cases with different levels of complexity prove that the MFBR method can effectively screen out the deformation region or gross error region, and the final registration results and deformation detection results also prove this.

5.2. Application Value and Limitations

The primary advantage of the MFBR method lies in its convenience and high level of automation. It does not require pre-established control points or manual selection of control points, nor does it necessitate the manual identification and extraction of stable areas [44]. Simply inputting DEM data of the region of interest for different periods enables rapid assessment of surface changes, particularly following landslide events. Moreover, the MFBR method can be applied to historical geomorphic data if such survey data are available. Importantly, however, some historical imagery may lack geographic information, unlike modern drone-based imagery. Therefore, the MFBR method uses the SURF operator to extract feature points for coarse registration, prioritizing simplicity and speed. With advancements in deep learning and artificial intelligence, new image feature matching techniques, such as SuperGlue and LoFTR, have emerged, which could improve image registration across different periods [45,46]. However, these methods typically require extensive training processes and the inclusion of multiple modules, which significantly increase the computational demand. This, in turn, presents challenges in terms of video memory usage and real-time operation. Balancing efficiency and accuracy remains an area that warrants further research.
The MFBR method can be influenced by various factors, such as large lakes, deserts, and other featureless areas, as well as regions with a very small proportion of stable areas. These conditions can lead to significant accuracy degradation or even failure of the method. However, this issue can be mitigated by expanding the data area. Moreover, the accuracy of the DEM data is also closely related to the results, which may be due to a variety of factors, such as image quality, terrain slope, and curvature [47,48,49]. Steep slopes and shadows occurring during image acquisition can result in stereo matching errors, thereby increasing DEM inaccuracies, which is a common challenge in photogrammetry [44]. In addition, the processing of vegetation, noise, and other factors during DEM processing contributes to the expansion of the stabilized area within the region and improves the accuracy of the method. Notably, in this study, the DEMs from different periods were resampled to the same resolution for registration and deformation calculations. While there is a small probability that some morphological information may be lost during this process, the results from the landslide studies in both regions demonstrate that the MFBR method can still achieve satisfactory registration and capture over 70% of the deformation information, even without additional refinement during data processing. In the future, the method’s applicability can be further extended by incorporating diverse monitoring data beyond DEMs. Additionally, computational efficiency can be enhanced through parallel computing and geocomputing technologies, aiming to improve both the speed and accuracy of landslide deformation monitoring.

6. Conclusions

In this work, to solve the problem of missing control points, a landslide deformation monitoring strategy based on the MFBR method is studied to register the multitemporal DEM and quantify deformation, and this strategy does not require ground support information. The 2D image features of the DEM are fully utilized by this strategy. Reliable feature points are extracted from different DEM images, which are not limited by the geographical location of the DEM and can effectively achieve rough registration of data. More importantly, in the precise registration stage, the possible deformation interference between multitemporal DEMs is considered, and the EM algorithm is used to pre-extract the stable region of the DEM to achieve accurate registration based on the stable region. Otherwise, the deformation region in the DEM will cause significant interference with the establishment of correspondence and may cause complete registration failure. Two realistic cases in the study show that the research method can accurately register the multitemporal DEM data of landslides and obtain deformation information. Moreover, the deformation monitoring ability can reach 83% of that with ground control points. This study provides a new idea and reference for landslide deformation monitoring without ground control points. Generally, large-scale survey data collected before a disaster do not specifically include ground control points. After a disaster, differences in geographic referencing may arise due to variations in operational departments, methods, and techniques. The proposed method enables rapid DEM data registration and deformation extraction, facilitating prompt monitoring of landslide conditions after an event and offering valuable data support for emergency response and rescue operations.

Author Contributions

All authors contributed significantly to this manuscript. Writing—original draft preparation, Y.W.; methodology, J.L. and P.D.; software, Y.W.; validation, X.Y.; field work, Y.W. and R.W. All authors have read and agreed to the published version of the manuscript.

Funding

This research was supported by the National Natural Science Foundation of China (No. 42361070, No. 41961061), the Yunnan Fundamental Research Projects (No. 202301AT070061, No. 202401AT070125), Revitalizing Yunnan Talents Support Program project funding support (Nos. YNWR-QNBJ-2020–048, YNWRQNBJ-2020–103), Yunnan Province Technical Innovation Talent Development Projects (No. 202405AD350058), Yunnan Province Innovation Team Project (202305AS350003), and the Yunnan Provincial Basic Research Project Key Project (No. 202201AS070024).

Data Availability Statement

The raw data supporting the conclusions of this article will be made available by the authors on request.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Froude, M.J.; Petley, D.N. Global fatal landslide occurrence from 2004 to 2016. Nat. Hazards Earth Syst. Sci. 2018, 18, 2161–2181. [Google Scholar] [CrossRef]
  2. Wu, C. Certainty Factor Analyses and Spatiotemporal Characteristics of Landslide Evolution: Case Studies in the Chishan River Watershed in Taiwan. ISPRS Int. J. Geo-Inf. 2022, 11, 382. [Google Scholar] [CrossRef]
  3. Iverson, R.M.; George, D.L.; Allstadt, K.; Reid, M.E.; Collins, B.D.; Vallance, J.W.; Schilling, S.P.; Godt, J.W.; Cannon, C.M.; Magirl, C.S.; et al. Landslide mobility and hazards: Implications of the 2014 Oso disaster. Earth Planet. Sci. Lett. 2015, 412, 197–208. [Google Scholar] [CrossRef]
  4. Intrieri, E.; Carlà, T.; Gigli, G. Forecasting the time of failure of landslides at slope-scale: A literature review. Earth-Sci. Rev. 2019, 193, 333–349. [Google Scholar] [CrossRef]
  5. Fang, K.; Tang, H.; Su, X.; Shang, W.; Jia, S. Geometry and Maximum Width of a Stable Slope Considering the Arching Effect. J. Earth Sci. 2020, 31, 1087–1096. [Google Scholar] [CrossRef]
  6. Yao, W.; Li, C.; Zuo, Q.; Zhan, H.; Criss, R.E. Spatiotemporal deformation characteristics and triggering factors of Baijiabao landslide in Three Gorges Reservoir region, China. Geomorphology 2019, 343, 34–47. [Google Scholar] [CrossRef]
  7. Wang, K.; Xu, H.; Zhang, S.; Wei, F.; Xie, W. Identification and Extraction of Geomorphological Features of Landslides Using Slope Units for Landslide Analysis. ISPRS Int. J. Geo-Inf. 2020, 9, 274. [Google Scholar] [CrossRef]
  8. Wu, Q.; Tang, H.; Ma, X.; Wu, Y.; Hu, X.; Wang, L.; Criss, R.; Yuan, Y.; Xu, Y. Identification of movement characteristics and causal factors of the Shuping landslide based on monitored displacements. Bull. Eng. Geol. Environ. 2018, 78, 2093–2106. [Google Scholar] [CrossRef]
  9. Lacroix, P.; Araujo, G.; Hollingsworth, J.; Taipe, E. Self-Entrainment Motion of a Slow-Moving Landslide Inferred From Landsat-8 Time Series. J. Geophys. Res. Earth Surf. 2019, 124, 1201–1216. [Google Scholar] [CrossRef]
  10. Charbonnier, S.; Deng, F.; Rodgers, M.; Xie, S.; Dixon, T.H.; Gallant, E.A.; López Vélez, C.M.; Ordoñez, M.; Malservisi, R.; Voss, N.; et al. High-resolution DEM generation from spaceborne and terrestrial remote sensing data for improved volcano hazard assessment—A case study at Nevado del Ruiz, Colombia. Remote Sens. Environ. 2019, 233, 111348. [Google Scholar]
  11. Tarolli, P.; Sofia, G.; Dalla Fontana, G. Geomorphic features extraction from high-resolution topography: Landslide crowns and bank erosion. Nat. Hazards 2010, 61, 65–83. [Google Scholar] [CrossRef]
  12. Leprince, S.; Barbot, S.; Ayoub, F.; Avouac, J.-P. Automatic and Precise Orthorectification, Coregistration, and Subpixel Correlation of Satellite Images, Application to Ground Deformation Measurements. IEEE Trans. Geosci. Remote Sens. 2007, 45, 1529–1558. [Google Scholar] [CrossRef]
  13. Zhang, D.; Yang, W.; Xu, C.; Ye, T.; Liu, Q. Extracting deforming landslides from time-series Sentinel-2 imagery. Landslides 2022, 19, 2761–2774. [Google Scholar] [CrossRef]
  14. Zhang, Y.; Li, Y.; Meng, X.; Liu, W.; Wang, A.; Liang, Y.; Su, X.; Zeng, R.; Chen, X. Automatic Mapping of Potential Landslides Using Satellite Multitemporal Interferometry. Remote Sens. 2023, 15, 4951. [Google Scholar] [CrossRef]
  15. An, P.; Fang, K.; Jiang, Q.; Zhang, H.; Zhang, Y. Measurement of Rock Joint Surfaces by Using Smartphone Structure from Motion (SfM) Photogrammetry. Sensors 2021, 21, 922. [Google Scholar] [CrossRef]
  16. Fang, K.; Dong, A.; Tang, H.; An, P.; Wang, Q.; Jia, S.; Zhang, B. Development of an easy-assembly and low-cost multismartphone photogrammetric monitoring system for rock slope hazards. Int. J. Rock Mech. Min. Sci. 2024, 174, 105655. [Google Scholar] [CrossRef]
  17. Chudý, F.; Slámová, M.; Tomaštík, J.; Prokešová, R.; Mokroš, M. Identification of Micro-Scale Landforms of Landslides Using Precise Digital Elevation Models. Geosciences 2019, 9, 117. [Google Scholar] [CrossRef]
  18. Mauri, L.; Straffelini, E.; Cucchiaro, S.; Tarolli, P. UAV-SfM 4D mapping of landslides activated in a steep terraced agricultural area. J. Agric. Eng. 2021, 52. [Google Scholar] [CrossRef]
  19. Eker, R.; Aydın, A.; Hübl, J. Unmanned aerial vehicle (UAV)-based monitoring of a landslide: Gallenzerkogel landslide (Ybbs-Lower Austria) case study. Environ. Monit. Assess. 2017, 190, 28. [Google Scholar] [CrossRef]
  20. Fan, J.-R.; Zhang, X.-Y.; Su, F.-H.; Ge, Y.-G.; Tarolli, P.; Yang, Z.-Y.; Zeng, C.; Zeng, Z. Geometrical feature analysis and disaster assessment of the Xinmo landslide based on remote sensing data. J. Mt. Sci. 2017, 14, 1677–1688. [Google Scholar] [CrossRef]
  21. Azmoon, B.; Biniyaz, A.; Liu, Z. Use of High-Resolution Multi-Temporal DEM Data for Landslide Detection. Geosciences 2022, 12, 378. [Google Scholar] [CrossRef]
  22. Pirasteh, S.; Shamsipour, G.; Liu, G.; Zhu, Q.; Chengming, Y.E. A new algorithm for landslide geometric and deformation analysis supported by digital elevation models. Earth Sci. Inform. 2020, 13, 361–375. [Google Scholar] [CrossRef]
  23. Li, J.; Wang, W.; Han, Z.; Li, Y.; Chen, G. Exploring the Impact of Multitemporal DEM Data on the Susceptibility Mapping of Landslides. Appl. Sci. 2020, 10, 2518. [Google Scholar] [CrossRef]
  24. Hu, S.; Qiu, H.; Pei, Y.; Cui, Y.; Xie, W.; Wang, X.; Yang, D.; Tu, X.; Zou, Q.; Cao, P.; et al. Digital terrain analysis of a landslide on the loess tableland using high-resolution topography data. Landslides 2018, 16, 617–632. [Google Scholar] [CrossRef]
  25. Hendrickx, H.; De Sloover, L.; Stal, C.; Delaloye, R.; Nyssen, J.; Frankl, A. Talus slope geomorphology investigated at multiple time scales from high-resolution topographic surveys and historical aerial photographs (Sanetsch Pass, Switzerland). Earth Surf. Process. Landf. 2020, 45, 3653–3669. [Google Scholar] [CrossRef]
  26. Haas, F.; Hilger, L.; Neugirg, F.; Umstädter, K.; Breitung, C.; Fischer, P.; Hilger, P.; Heckmann, T.; Dusik, J.; Kaiser, A.; et al. Quantification and analysis of geomorphic processes on a recultivated iron ore mine on the Italian island of Elba using long-term ground-based lidar and photogrammetric SfM data by a UAV. Nat. Hazards Earth Syst. Sci. 2016, 16, 1269–1288. [Google Scholar] [CrossRef]
  27. Peppa, M.V.; Mills, J.P.; Moore, P.; Miller, P.E.; Chambers, J.E. Automated co-registration and calibration in SfM photogrammetry for landslide change detection. Earth Surf. Process. Landf. 2018, 44, 287–303. [Google Scholar] [CrossRef]
  28. Zhang, T.; Cen, M. Robust DEM co-registration method for terrain changes assessment using least trimmed squares estimator. Adv. Space Res. 2008, 41, 1827–1835. [Google Scholar] [CrossRef]
  29. Chen, X.; Cen, M.; Guo, H.; Zhang, T.; Zhao, C.; Zhang, B. Chinese satellite photogrammetry without ground control points based on a public DEM using an efficient and robust DEM matching method. Int. J. Remote Sens. 2017, 39, 704–726. [Google Scholar] [CrossRef]
  30. Noh, M.-J.; Howat, I.M. Automated Coregistration of Repeat Digital Elevation Models for Surface Elevation Change Measurement Using Geometric Constraints. IEEE Trans. Geosci. Remote Sens. 2014, 52, 2247–2260. [Google Scholar] [CrossRef]
  31. Nuth, C.; Kääb, A. Co-registration and bias corrections of satellite elevation data sets for quantifying glacier thickness change. Cryosphere 2011, 5, 271–290. [Google Scholar] [CrossRef]
  32. Karkee, M.; Steward, B.L.; Aziz, S.A. Improving quality of public domain digital elevation models through data fusion. Biosyst. Eng. 2008, 101, 293–305. [Google Scholar] [CrossRef]
  33. Besl, P.J.; McKay, N.D. A Method for Registration of 3-D Shapes. In Sensor Fusion IV: Control Paradigms and Data Structures; SPIE: Bellingham, WA, USA, 1992; Volume 1611, pp. 586–606. [Google Scholar]
  34. Torlegard, D.R.K. Three-dimensional absolute orientation of stereo models using digital elevation models. Photogramm. Eng. Remote Sens. 1988, 54, 1385–1389. [Google Scholar]
  35. Pilgrim, L. Robust estimation applied to surface matching. ISPRS J. Photogramm. Remote Sens. 1996, 51, 243–257. [Google Scholar] [CrossRef]
  36. Tsunetaka, H.; Hotta, N.; Hayakawa, Y.S.; Imaizumi, F. Spatial accuracy assessment of unmanned aerial vehicle-based structures from motion multi-view stereo photogrammetry for geomorphic observations in initiation zones of debris flows, Ohya landslide, Japan. Prog. Earth Planet. Sci. 2020, 7, 24. [Google Scholar] [CrossRef]
  37. Moon, T.K. The expectation-maximization algorithm. IEEE Signal Process. Mag. 1996, 13, 47–60. [Google Scholar] [CrossRef]
  38. Koch, K.R. Robust estimation by expectation maximization algorithm. J. Geod. 2012, 87, 107–116. [Google Scholar] [CrossRef]
  39. Sammaknejad, N.; Zhao, Y.; Huang, B. A review of the Expectation Maximization algorithm in data-driven process identification. J. Process Control 2019, 73, 123–136. [Google Scholar] [CrossRef]
  40. Levoy, S.R.M. Efficient variants of the ICP algorithm. In Proceedings of the Third International Conference on 3-D Digital Imaging and Modeling, Quebec City, QC, Canada, 28 May–1 June 2001; pp. 145–152. [Google Scholar]
  41. Lane, S.N.; Westaway, R.M.; Murray Hicks, D. Estimation of erosion and deposition volumes in a large, gravel-bed, braided river using synoptic remote sensing. Earth Surf. Process. Landf. 2003, 28, 249–271. [Google Scholar] [CrossRef]
  42. Neugirg, F.; Stark, M.; Kaiser, A.; Vlacilova, M.; Della Seta, M.; Vergari, F.; Schmidt, J.; Becht, M.; Haas, F. Erosion processes in calanchi in the Upper Orcia Valley, Southern Tuscany, Italy based on multitemporal high-resolution terrestrial LiDAR and UAV surveys. Geomorphology 2016, 269, 8–22. [Google Scholar] [CrossRef]
  43. Hodson, T.O. Root-mean-square error (RMSE) or mean absolute error (MAE): When to use them or not. Geosci. Model Dev. 2022, 15, 5481–5487. [Google Scholar] [CrossRef]
  44. Maurer, J.; Rupper, S. Tapping into the Hexagon spy imagery database: A new automated pipeline for geomorphic change detection. ISPRS J. Photogramm. Remote Sens. 2015, 108, 113–127. [Google Scholar] [CrossRef]
  45. Sarlin, P.-E.; DeTone, D.; Malisiewicz, T.; Rabinovich, A. SuperGlue: Learning Feature Matching with Graph Neural Networks. In Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, 14–19 June 2020; pp. 4937–4946. [Google Scholar]
  46. Bao, H.; Zhou, X.; Sun, J.; Shen, Z.; Wang, Y. LoFTR:Detector-FreeLocalFeatureMatchingwithTransformers. In Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Nashville, TN, USA, 19–25 June 2021; pp. 8922–8931. [Google Scholar]
  47. Carlisle, B.H. Modelling the Spatial Distribution of DEM Error. Trans. GIS 2005, 9, 521–540. [Google Scholar] [CrossRef]
  48. Höhle, J.; Höhle, M. Accuracy assessment of digital elevation models by means of robust statistical methods. ISPRS J. Photogramm. Remote Sens. 2009, 64, 398–406. [Google Scholar] [CrossRef]
  49. Gardelle, J.; Berthier, E.; Arnaud, Y. Impact of resolution and radar penetration on glacier elevation changes computed from DEM differencing. J. Glaciol. 2017, 58, 419–422. [Google Scholar] [CrossRef]
Remotesensing 16 04236 g001
Figure 1. Workflow of the MFBR method.
Figure 1. Workflow of the MFBR method.
Remotesensing 16 04236 g001
Remotesensing 16 04236 g002
Figure 2. Image-based DEM position registration method.
Figure 2. Image-based DEM position registration method.
Remotesensing 16 04236 g002
Remotesensing 16 04236 g003
Figure 3. 3D spatial-feature-based precision registration method.
Figure 3. 3D spatial-feature-based precision registration method.
Remotesensing 16 04236 g003
Remotesensing 16 04236 g004
Figure 4. Simulation dataset.
Figure 4. Simulation dataset.
Remotesensing 16 04236 g004
Remotesensing 16 04236 g005
Figure 5. Luchun landslide and data collection.
Figure 5. Luchun landslide and data collection.
Remotesensing 16 04236 g005
Remotesensing 16 04236 g006
Figure 6. The Gongshan landslide.
Figure 6. The Gongshan landslide.
Remotesensing 16 04236 g006
Remotesensing 16 04236 g007
Figure 7. Deformation detection results of the simulation data. (a) Raw deformation. (b) Based on the MFBR method.
Figure 7. Deformation detection results of the simulation data. (a) Raw deformation. (b) Based on the MFBR method.
Remotesensing 16 04236 g007
Remotesensing 16 04236 g008
Figure 8. Deformation detection results for the Luchun landslide. (a) Based on GCPs. (b) Based on MFBR. (c) ICP-based method.
Figure 8. Deformation detection results for the Luchun landslide. (a) Based on GCPs. (b) Based on MFBR. (c) ICP-based method.
Remotesensing 16 04236 g008
Remotesensing 16 04236 g009
Figure 9. Deformation detection results for the Gongshan landslide. (a) Based on GCPs. (b) Based on MFBR. (c) ICP-based method.
Figure 9. Deformation detection results for the Gongshan landslide. (a) Based on GCPs. (b) Based on MFBR. (c) ICP-based method.
Remotesensing 16 04236 g009
Remotesensing 16 04236 g010
Figure 10. Stable region extraction results.
Figure 10. Stable region extraction results.
Remotesensing 16 04236 g010
Table 1. Registration accuracy/m.
Table 1. Registration accuracy/m.
Raw DataMFBRICP
Register (RMSE)/0.96312.189
Check points (Mean)0.4611.3683.754
Check points (RMSE)0.5522.5284.617
Table 2. Registration accuracy/m.
Table 2. Registration accuracy/m.
DataRegister Accuracy (MFBR)Register Accuracy (ICP)
Luchun landslide0.3680.745
Gongshan landslide2.4563.258
Table 3. Summary of the deformation detection capabilities.
Table 3. Summary of the deformation detection capabilities.
Methodsab(d)
Luchun landslideMFBR83.54%
ICP60.69%
Gongshan landslideMFBR71.24%
ICP47.71%
Simulation dataMFBR73.90%
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Wang, Y.; Li, J.; Duan, P.; Wang, R.; Yu, X. The DEM Registration Method Without Ground Control Points for Landslide Deformation Monitoring. Remote Sens. 2024, 16, 4236. https://doi.org/10.3390/rs16224236

AMA Style

Wang Y, Li J, Duan P, Wang R, Yu X. The DEM Registration Method Without Ground Control Points for Landslide Deformation Monitoring. Remote Sensing. 2024; 16(22):4236. https://doi.org/10.3390/rs16224236

Chicago/Turabian Style

Wang, Yunchuan, Jia Li, Ping Duan, Rui Wang, and Xinrui Yu. 2024. "The DEM Registration Method Without Ground Control Points for Landslide Deformation Monitoring" Remote Sensing 16, no. 22: 4236. https://doi.org/10.3390/rs16224236

APA Style

Wang, Y., Li, J., Duan, P., Wang, R., & Yu, X. (2024). The DEM Registration Method Without Ground Control Points for Landslide Deformation Monitoring. Remote Sensing, 16(22), 4236. https://doi.org/10.3390/rs16224236

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop