Land Movement Detection from UAV Images for a Sustainable World

a steep hillside in the Reina del Cisne sector (SE Cuenca) affecting several homes, providing recommendations and future work directions.


METHODS
Figure 1.Study area (Cuenca -Ecuador) with information about the landslide and distribution of the profiles studied

Generation of point clouds using UAV
The UAV used was a DJI Phantom 4 RTK®.This UAV is ideal for mapping, inspection, and surveying, featuring a high-resolution camera capable of capturing RTK data with centimeter-level accuracy and requiring fewer ground control points compared to traditional tools (Peppa et al., 2019).The UAV can produce orthomosaics, point clouds, and digital elevation models (DEM) with a precision of 3 cm (Mulakala, 2019).
The study area was delimited using PPGIS (Public Participation Geographic Information System). Figure 1 shows the study area located at UTM coordinates 17 S area, 726.775 m E and 9.679.327m N, with an elevation of 2600 m, characterized by steep slopes.
Landslides are among the most destructive geological processes affecting humans, causing deaths and property losses worth tens of trillions of dollars each year (Brabb & Harrod, 1989).In Cuenca-Ecuador, many inhabitants build their homes on steep slopes due to the low prices of such lands and the rapid urban growth.These slopes often experience landslides, which damage constructions, as seen in the Reina del Cisne sector.It is essential to monitor these slides for accurate mapping, urban planning, and risk prevention.Monitoring can be done using classical in-situ topographic techniques such as differential GNSS or total station (González-Zúñiga, 2010), combined with active sensors like RADAR (Bardi et al., 2014;Martire et al., 2016;Ventisette et al., 2014), UAVs (Dewitte et al., 2008;Martínez-Espejo Zaragoza et al., 2017;Buffi et al., 2017), or satellite images (Behling & Roessner, 2017).Detecting large landslides is challenging, requiring significant economic and human resources and long periods, typically weeks or months.Changing climatic conditions, such as heavy rains, also cause morphological variations.In Ecuador, heavy rains often lead to landslides, necessitating improved detection processes for at-risk areas.This study aims to monitor and record sliding movements on a steep hillside in the Reina del Cisne sector (SE Cuenca) affecting several homes, providing recommendations and future work directions.
The flight was configured using the Pix4capture© mobile application in double grid flight mode for 3D models (Figure 2).

Points
Error (mm) The two sets of images from January 2019 and June 2019 were processed using Agisoft PhotoScan© software.The processing steps included: 1) photograph selection; 2) camera calibration; 3) finding homologous points; and 4) identifying points over the terrain.The UAV made real-time corrections and triangulations to adjust the images, generating point clouds, orthophotos, and DEM (Núñez Calleja, 2016).
After processing, two point clouds were obtained with densities shown in Table 2.

Errors obtained in the point clouds generation:
The joining of scans to form each of the 2 points clouds was satisfactory.The highest average error for the two dates was only 2.3 cm (Figure 3).
To align and compare in CloudCompare the point clouds of both dates (January 2019 and June 2019) were used 12 static reference points, located outside the landslide area indicated in Figure 4.In addition, a comparison has been made of the deformations that the dwelling has suffered using the same 12 reference points located outside the main escarpment (January 2019 and June 2019).

Errors obtained in the point clouds alignment:
The 12 reference points were precisely located and selected in each of the point clouds to align them.The points (A) correspond to the January point cloud and the points (R) correspond to the June point cloud, for the first alignment, which is indicated in Figure 4.
The alignment errors were minimal, with all errors below 1 mm, as shown in Table 4.
In Profile 1, (Figure 15) and Profile 3, (Figure 17) are displayed along the same as the recorded terrain movements, although they are removed from the central axis of the landslide these are produced by mini-slides produced by secondary escarpments.
In Profile 2, (Figure 16), it can be noticed that the recorded movements are greater, this is because the profile is located in the central axis of the landslide, besides being influenced by the movements of the mini-landslide produced by a secondary escarpment.-: Sinking.

CONCLUSIONS
The segmentation method for scanning profiles was used for comparing profiles by segmenting the point cloud along lines.This technique involves freely drawing a profile across the points.The tracing of this line can be done where it is most convenient (Gonzalez et al., 2004), in this case, along the terrain where the landslide showed the most activity during field visits.The profile representation was generated using AutoCAD® Civil 3D with the extracted profiles from the point clouds (January 2019 and June 2019).This provided profiles with relevant landslide movement information.

Procedure with extracted profiles
The method for comparing profiles (January 2019 and June 2019) involved exporting the point clouds to AutoCAD® Civil 3D, creating a surface from the point cloud, drawing a line along the zone of interest at the top of the point cloud, creating a profile of the surface corresponding to the line, and extracting and displaying the abscissas and heights information for the profile.Figures 12, 13, and 14 show the three January 2019 point cloud profiles.This study successfully demonstrated the application of UAV photogrammetry and point cloud analysis to detect and monitor landslides in the Reina del Cisne sector of Cuenca, Ecuador.
The main conclusions drawn from this research are as follows: The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLVIII-2-2024 ISPRS TC II Mid-term Symposium "The Role of Photogrammetry for a Sustainable World", 11-14 June 2024, Las Vegas, Nevada, USA

Effectiveness of UAV Photogrammetry:
The use of a DJI Phantom 4 RTK® UAV provided highresolution imagery that facilitated the generation of accurate point clouds and digital elevation models (DEMs).
The UAV method proved to be efficient and cost-effective, reducing the need for extensive ground control points and providing rapid data acquisition suitable for landslide monitoring.

Point Cloud Comparison:
By comparing point clouds generated from UAV images taken in January 2019 and June 2019, significant land movements were detected.The high density of points (78 million in January and 75 million in June) ensured detailed analysis.
The alignment and profiling of these point clouds using CloudCompare and AutoCAD® Civil 3D allowed for precise quantification of deformations and provided valuable insights into the extent and progression of the landslide.

Landslide Activity Detection:
The study identified critical zones of movement and deformation on the hillside, indicating ongoing landslide activity.The profiles extracted showed notable displacements, which were crucial for understanding the dynamics of the landslide.
The methodology highlighted the potential for early warning systems, as regular UAV surveys can detect changes over time, allowing for timely interventions and risk mitigation.

Implications for Urban Planning and Safety:
The findings underscore the importance of integrating advanced monitoring techniques in urban planning, especially in areas prone to landslides.This can lead to better-informed decisions, safer construction practices, and enhanced disaster preparedness.
The study also emphasizes the need for continuous monitoring and the establishment of early warning systems to protect inhabitants and minimize property damage.

Future Work:
Future research should focus on extending the monitoring period and incorporating additional data points to improve the accuracy of predictions.
Integrating other remote sensing technologies, such as LiDAR and InSAR, could provide complementary data to enhance the understanding of landslide mechanisms.
There is also a need to develop automated systems for real-time data processing and analysis, facilitating quicker response times and more effective landslide management.
In conclusion, UAV photogrammetry combined with point cloud analysis presents a robust tool for landslide detection and monitoring, offering significant benefits for sustainable urban development and disaster risk reduction.

Figure 3 .
Figure 3. Average errors in m the generation of points clouds

Table 2 .
Point clouds density

Table 3 .
Error values in the point clouds generation

Table 4 .
Error values in the alignment between January 2019 and June 2019 clouds