There is a three-stage process for the images provided by the UAV. The stages are 3D cloud formation, georeferencing, and point cloud comparison. The formation of a 3D point cloud is a process that involves different steps. Some software packages can be utilized for these steps like PhotoScan (Agisoft) and PhotoModeler as well as open-source platforms such as Photosynth and Bundler (Harwin & Lucieer (2012), Snavely et al., 2010). For georeferencing and comparison of point clouds from various epochs, software such as CloudCompare, I-Site Studio, and 3DReshaper are perfectly useful. These software packages actually allow point cloud comparison by aligning the different epochs using existing reference points in different 3D models (Vazaios et al., 2017, Lato & Vöge (2012), Dewez et al., 2016).
In this study, the photogrammetry process follows different steps as shown in Fig. 1. Following the flowchart, begin with, first of all, a walkover study was conducted in the location and after doing a desk study, the topography data of the study area was surveyed. After that, the location of the targets was schematized due to the initial data analyses. The retroreflective targets were printed on the A4-sized paper and later laminated to provide more resistance to climate conditions. Placing the targets on-site facilitated easy alignment from one epoch to another. On the other hand, the location of the targets was heavily dependent on the topography of the study area so the location, as well as the quantities, would be different in various areas. The targets were moreover utilized as ground control points, and the arrangement of the points has been recorded with a help of the Pentax Arrangement G6 GNSS. This array of ground control points (known points) was used in order to improve the precision of the point cloud or the digital elevation model (DEM) produced utilizing the UAVs.
In the current study, a UAV (DJI PHANTOM 3 Pro) was utilized which had been equipped with a 12-megapixel camera (Sony EXMOR). It provided 120 pictures of every epoch remotely. All the pictures collected from the UAV were supposed to overlap each other by 70%. Regarding the intact photogrammetry, the battery life and flashcard memory space were continuously checked.
The software utilized for image processing was PhotoScan. The advantage of using this software is that it provides an affordable solution for multi-view 3D construction. Structure-from-motion (SfM) photogrammetry gives hyper-scale three-dimensional (3D) landform models utilizing overlapping images obtained according to various points of view (Eltner & Sofia (2020), Santagati (2013)). At the site, estimations were taken to guarantee that the above standards were met and around 120 photos were taken at every epoch. Captured images were in .jpg format and their sizes were 4000 × 3000 pixels. The images were stored on a video speed class (V90) micro-SD card to enable fast recording. Pictures were straightforwardly imported from the micro-SD card to the PC with a DJI link cable. After downloading the images from the memory card, it is necessary to delete the dislocated and blurred pictures to obtain a better quality of the 3D point cloud construction. In this study, camera calibration was not necessary due to bundle adjustment (automatic estimation of camera calibration parameters using Brown’s model (1966) for lens distortion) and the use of standard optical lenses and a highly redundant photo network.
After loading the taken photographs into PhotoScan, the images are aligned. This cycle, which takes 10 minutes, iteratively refines the inner just as outer camera orientations and areas utilizing the least-squares arrangement. In this stage, the software builds a sparse point cloud model and calculates the depth information based on the estimated camera positions. In the software, the command 'Build dense point cloud' can produce a single dense point. Moreover, the software operator can set the quality and depth for the point cloud generation. The computer configuration utilized in this study is an i7- 7700HQ CPU at 2.8 GHz with 16 GB of RAM and a K5000 graphics card and a 1 TB SSD hard drive.
For data processing, first of all, it was needed to export the generated point clouds, which were obtained from two different epochs, to the CloudCompare software in the LAZ (Lidar Data file) format. CloudCompare is open source 3D point cloud handling software. It has been initially intended to play out a correlation between two dense 3D points clouds. Before doing any comparison, the Iterative Closed Point Processing algorithm of the CloudCompare software aligns the two point clouds. It is important to consider the Close overlap between the targets to produce the best alignment of the point clouds. It has been revealed that noise and points outside the intended space ought to be taken out prior to playing out the alignment and enrollment to forestall the degradation of the enlisted point clouds.
Change identification alludes to the method involved with distinguishing the distinctions in an object by noticing it at two epochs. The regions in the point cloud where changes happened were investigated all the more intently while carrying out change detection (Sinha et al., 2010).
Monitoring of the retaining wall of a slop beside the Kyrenia Castle
In this case study, it was tried to monitor the retaining wall of a slope located beside the Kyrenia Castle in the northern part of Cyprus as it has shown if Fig. 2 via close-range photogrammetry. As this place is located adjacent to the castle and the sea, it is really important to protect it not only because of the safety of the castle but it also important via environmental problems. The author was responsible to do an investigation via the Kyrenia municipality organization announcement about the huge erosion of the intended slope and its retaining walls.
There are two FEM software which are geo5 and plaxis 2D were used in this study to analyze the area in 2D dimension. The results showed that the observed slide has both rotational and transitional features. The other main part that was revealed in the investigation via the FEM tools was the presence of the toe of the slide in the sea. As it is indicated in Fig. 3, the footing of the retaining walls are not passes the slide and this occurrence would load up the slide and speed up the destructive process. The movement vectors in the photo express the direction of the slide, and as can be seen there is no force there to resist the movement. From the image, it can be seen that the toe is located in the sea, and this is really important for designing the protection procedure.
In the photogrammetry process, first of all, the satellite photos of the area from 2008 to 2020 were obtained from google earth. As can be seen in Fig. 4, in 2008 there had not been any significant problem with the slope and therefore there seems to be nothing wrong with the slope; however, in 2010 the landslide started primarily and the tip was also obvious, as it was shown in Fig. 4. After three years that the destructive process of the landslide was continuing, the organization of Kyrenia Municipality decided to construct retaining walls to protect the slope in 2013, as is obvious from the related photo in Fig. 4. Due to the lack of design, the footing of the walls is embedded over the slide instead of passing through it. Therefore, these walls not only protect the slope from sliding but also aggravated it regarding their undesirable load. The destructive process has been continuing until 2020, which is indicated in the last photo can endorse while the half of car park in the picture has been destroyed in comparison with the photo in 2008.
In the photogrammetry process using an attached camera to the UAV a dense cloud image has been prepared, which can be seen in Fig. 5. The photogrammetry process has been accomplished over a period of 2 years.
This model is also known as Digital Twin, represents the virtual copy of the physical object. The Digital Twin duplicates the exact properties and behaviors of its physical counterpart in the physical space through modeling and real-time data exchange, allowing for learning, reasoning, and dynamically re-calibrating for improved decision-making (Glaessgen & Stargel (2012), Grieves (2014)).
As each pixel of the photo in Fig. 5 contains geometric data, the results of comparing these data in different epochs revealed the movement process of the wall in the specific section indicated in Fig. 5(A-A).
As it can be seen from Fig. 6, the movement of the wall from January 2018 to December 2019 occurred in different time periods. The significant displacement occurred between epoch-1 and epoch-2 and the value is about 50 cm, which is really high.
As it is indicated in the numerical study part, the toe of the slide was expected to be placed in the sea. To clarify the mentioned point as much as possible, a bathymetric survey was adopted. The bulge area was surveyed by a GNSS receiver and the topography results have been proved that the toe of the slide is in the water as it can be seen in Fig. 7. In the end for checking out the accuracy of the method, the results could be compared with the inclinometer, GNS, or laser scanner results.