The current technique used to measure construction is the conventional total station method. However, the conventional method is time-consuming and could not be used to create a photo-realistic three-dimensional (3D) model of an object. Furthermore, the Canseleri building is located at a slope. The paper aims to discuss this issue.
The aim of this study is to assess the geometric accuracy of a 3D model using unmanned aerial vehicle (UAV) images. There are two objectives in this study. The first is to construct a 3D model of the Canseleri building using UAV images. The second objective is to validate the 3D model of the Canseleri building based on actual measurements.
The close-range photogrammetry method, using the UAV platform, was able to produce a 3D building model. The results show that the errors between the actual measurement and the generated 3D model were less than 4 cm. The accuracy of the 3D model achieved in this study was about 0.015 m, compared to total station measurements.
Accuracy assessment was done by comparing the estimated measurement of the 3D model with the direct measurement. The differences between the measured values with actual values could be compared. Based on this study, the 3D building model gave a reliable accuracy for specific applications.
Ahmad Shazali, A. and Tahar, K. (2019), "Virtual 3D model of Canseleri building via close-range photogrammetry implementation", International Journal of Building Pathology and Adaptation, Vol. 38 No. 1, pp. 217-227. https://doi.org/10.1108/IJBPA-02-2018-0016Download as .RIS
Emerald Publishing Limited
Copyright © 2019, Emerald Publishing Limited
Photogrammetry is a technique developed from the advancement of the science of measuring object dimensions in photographs (Mölg and Bolch, 2017). The creation of a photo-realistic 3D model is popular in geomatics society nowadays. Mesas-Carrascosa et al. (2018) stated that the advancement of technology such as faster computer processing units increases the ability of computers to render a high-quality 3D model. The CRP technique creates many applications of 3D models, such as 3D games, cultural heritage documentation, virtual reality, special effects and geospatial application (Counts et al., 2016; Guidi et al., 2014). The images captured require further processes using related Photogrammetry software tools such as Agisoft PhotoScan to create a 3D model.
The need for 3D models is growing and expanding rapidly in various fields, especially in architecture. The most popular techniques in creating 3D models are using a laser scanner and the photogrammetry method. A laser scanner is a well-known albeit expensive technology in creating 3D models (Abu Hanipah and Tahar, 2018). Close-range photogrammetry (CRP) is a combination of art, science and technology used to obtain precise mathematical measurements and 3D data from two or more photographs (Ginovart et al., 2014). It is a measurement technique that calculates the 3D coordinates of an object out of the measurement of two or more images of the object from different angles of positions. Brunier et al. (2016) mentioned that the CRP is less expensive, but the quality of the output is not as high as using the laser scanner.
The CRP technique uses a Digital Single Lens Reflex for image collection (Stöcker et al., 2015). The Department of Civil Engineering in the Indian Institute of Technology Roorkee did a 3D model of a massive object and successfully created a virtual 3D photo-realistic textured model of its Department of Civil Engineering building. The authors of this study claimed that the quality of the model depended on the camera quality, while the accuracy depended on various matters, such as camera calibration, camera resolution, the geometry of camera position and marking of features on the image (Heilman et al., 2018).
Creating a 3D model by the utilisation of 2D images is known as image-based modelling. 3D models are used in a variety of fields other than geomatics such as the medical industry, as the CRP technique is used to model organs. 3D models are also used by the movie and gaming industries for animation purposes (Perez and Robleda, 2015). Architects use 3D models to propose their virtual building and landscape. Engineers use them to design new vehicles and structures (Hoang, 2018). The CRP technique visualises more realistic models compared to graphic-based object models. The conventional CRP technique uses a digital camera as a tool to capture the images of an object (Brunier et al., 2016). This conventional technique can only access small or human-size objects.
According to Remondino (2014), unmanned aerial vehicle (UAV) photogrammetry indeed promotes various new applications in the close-range aerial technique, while introducing low-cost alternatives to the classical manned aerial photogrammetry. This can be explained by the use of low-cost platforms combined with the use of amateur digital cameras installed with GPS/GNSS systems. The GPS/GNSS system is used to navigate the UAV with high precision to capture images based on waypoints that are widely used. The major advantage of the UAV system over the traditional manned airborne system is its high flexibility that allows image acquisition from different points of view with lower costs compared with classical aerial photogrammetry as a platform (Nex and Remondino, 2014).
Furthermore, the UAV allows high-resolution images to be obtained. If the survey site is small and the use of a manned aircraft to capture aerial photos is not economical, the use of UAV platforms is very suitable. The variety of digital cameras on the market has strongly enhanced the capabilities of UAV. The UAV is a new technique introduced to create the 3D model of a larger-scale object. The UAV platform can cover high rise buildings and a large field area. This is a new technology that requires further skills to pilot a drone and capture images. Simultaneously, accuracy is the main concern in a photogrammetric survey. Thus, a comparison will be made between the measurements that use CRP and a building plan that is measured accurately (Peppa et al., 2016).
One of the disadvantages of UAV is the payload. Low-weight sensors, such as small- or medium-size amateur cameras, are good alternatives as photogrammetric measurement systems. Furthermore, UAVs are equipped with low-cost sensors that reduce the quality in the mapping data, position of the systems and the orientation (Genesio et al., 2015). UAVs are highly dependent on environmental conditions, such as wind due to its lightweight. Each image should have the same lighting conditions during image acquisition to get uniform illumination in each image.
UAV has been used for heritage documentation due to its accuracy and ability to record building texture information. A number of case studies have reported the documentation and restoration of heritage buildings, such as agro-industrial buildings, a historic castle and a fire-damaged historical building. Digital 3D models were constructed using the CRP data, which were used to restore the buildings. El Meouche et al. (2016) researched the use of a UAV to mount a CRP system to survey the Tuekta burial mounds in the Russian Altay. The RMS error achieved was within 0.077–0.082 m.
Several processes need to be carried out to develop the reconstruction of a 3D model using the UAV platform. One of the UAV products known as the DJI Phantom 3 Standard was used in this study. The research design was divided into four phases: namely planning, data acquisition, data processing and data analysis (Figure 1).
Location of study
The study was conducted in Universiti Teknologi MARA (UiTM), Shah Alam, Malaysia. The building that was chosen to create a 3D model was the Canseleri building that is located near the main gate of UiTM. The iconic Canseleri building is the main administration building of UiTM and is one of the biggest buildings in UiTM. The size of the building was one of the challenges of creating a 3D model of it. Careful planning was needed to acquire the building’s images. The building itself was located at a slope that made it challenging to perform the CRP work. Plate 1 shows the building of interest in this study.
In a survey of work practice, the first and most important step is planning. In planning, reconnaissance work is carried out and many things discovered. Narrow spaces around the Canseleri building did not enable the UAV to cover those areas during image acquisition. On the first day, the UAV, which was the DJI Phantom 3 Standard, was flown to test the image acquisition process. The drone used could not fly more than 100 metres above the Canseleri building due to the limitation of the DJI Phantom 3 Standard. As a result, the UAV’s camera did not capture images of several areas. Furthermore, the rooftop of the Canseleri building was filled with aerials and antennas, causing the drone to lose signal due to magnetic distortion. The area of survey work needed to be visited and identified first.
Images of the Canseleri building were acquired from the UAV’s (DJI Phantom 3 Standard) integrated camera. Planning of autonomous and manual flight missions is important as good planning can increase work efficiency. Automatic flight planning can be done using software or apps such as DroneDeploy. Another design for manual image capturing is the oblique method (Cavegn et al., 2014). The planning considered different aspects, such as the size of the site, the shape of the object on site, and the height of the structure or object itself. Time and weather are also considered in flight planning. Since the Canseleri building’s surrounding area is covered by high trees and limited in space, the pilot had to manually manoeuvre the UAV to capture the images around the building. The altitude must be maintained during the image capturing process. There were standards set up that considered the wind, cruise-speed, exposure, white balance and other camera parameters. The flights were planned to make sure that the overlap between the photos must be between 65 and 85 per cent to produce an accurate 3D model. The waypoints of the UAV were planned in order to avoid any obstacle during image acquisition. Three different levels of altitude of the oblique method were planned to cover all sides of the building (Figure 2). Figure 2 shows the positions of cameras during image acquisition. The camera position covered the whole Canseleri building from different altitudes.
The highest level of image resolution was set at 4,000 × 3,000 pixels to achieve high-quality textures. Since the study focussed on the massive Canseleri building, a total of 118 images were taken from three different altitudes for CRP modelling to cover the whole structure. The three different altitudes were necessary to increase image redundancies during image processing. Ideally, the ratio of the distance between cameras and the target object should be within a reasonable limit. The values of this ratio should be between 1/15 and 1/20 (Singh et al., 2013). The images were acquired from the three altitudes to ensure a high-quality 3D model and avoid any “gaps” resulting from the lack of photo coverage. In this study, the camera angle during image acquisition was about 45° (Vacca et al., 2017). Figure 3 shows the angle of the camera during image capturing.
The main purpose of camera calibration was to determine the camera parameters, such as the CCD format size of a digital camera (Fw and Fh), lens distortion (k1, k2, k3, and p1 and p2), the principle point (Xp and Yp), and the camera focal length (f). The calibration was required to correct the internal geometry of each image based on camera parameters. Figure 4 shows the example of radial lens distortion of a single image. The radial lens distortion parameters from the camera calibration could be used to correct the radial lens distortion during image processing. The final calibration report can be obtained after the calibration process is completed.
Automatic calibration process can be done on the photogrammetry software. All images were uploaded to the Agisoft PhotoScan software where the auto-marking function detected all target points and adjusted the camera parameters, such as lens distortion and focal length. All images were processed on the software self-calibration tools where any distortions on the images were determined and used for image processing. The true measurements were necessary to calculate the accuracy of the 3D model that was generated compared to the actual building. In this study, true measurements were the direct measurement of the building parts, such as windows, entrance and pillars. The accuracy of the true measurement values allowed the assessment of the measurement of the 3D model of the Canseleri building. The measurements were done using a total station to get the final true measurement value. Each measurement was observed at least three times to make sure it was free from any errors. Plate 2 shows the direct measurement works for the 3D model accuracy assessment
Data processing and analysis
The next step after image acquisition and camera calibration was image processing. The processes in Agisoft PhotoScan software included 3D modelling and accuracy measurement. The processing steps are described in Figure 5. The main data in this study were the images taken from the UAV in order to create the 3D model. It was necessary to use only good photos for 3D model construction before starting any operation. Good photos refer to images that are sharp and not blurry. The photos needed to be aligned once uploaded in the software. In this stage, the image processing was done using the align photos tool. Here, photo stitching was carried out where the image matching process was done by the computer. The output of aligning photos depended on the computed camera position and a sparse point cloud. The alignment result could be inspected and any incorrectly positioned photos could be removed. Matches between any two points could be seen using the view matches tool. Following the workflows in Agisoft PhotoScan, the next step after photo-aligning was building the dense point cloud from the images.
Agisoft PhotoScan allowed the generation and visualisation of a dense point cloud model. The camera positions were estimated, and the programme calculated the depth information for each camera angle. Subsequently, the depth information was combined in a single dense point cloud. Building mesh required a lot of processing time, depending on the specification of the computer used. Mesh was crucial for the next stage, namely building the object texture. In this stage, triangles formed the shape of the 3D object. Photoscan supported several reconstruction designs and settings, depending on the objective of the final output. The reconstruction parameters also depended on the given data set to produce optimal output. These parameters are surface type, source data and face count, also known as polygon count.
The texture mapping mode determined the formation of the object texture in the texture atlas. Optimal texture packing could be achieved by selecting the proper texture mapping mode to get the best visual of the final 3D model. Texture size specified the size that are the width and height of the texture atlas in pixels, and processed the number of files for texture. A greater resolution of the final model could be achieved by exporting the texture into multiple files. The risk of failure during processing due to the high resolution might occur because of RAM limitation. There is also an additional parameter, the enable colour correction setting.
Results and analysis
The calibration of the camera was vital as the images could affect the quality and accuracy of the 3D model. There were several ways to calibrate a camera using the Agisoft PhotoScan that did not require manual calibration. There was a built-in calibration in the software that could automatically calibrate the images captured by the camera or known as self-calibration. After undergoing several processes, the 3D model of the Canseleri building was successfully created. Once the photos were loaded into the software, they must be aligned. At this stage, the Agisoft PhotoScan performed the camera orientation with the position of each photo. After the computer processed the image alignment, the position of the photos was revealed and the sparse point cloud model was displayed. The quality of the 3D model was based on the quality of the images from the camera and the specifications of the computer used to process the 3D model. The higher the quality of the images, the higher the quality of the 3D model. A computer with high specifications could process the photos precisely with fine details, thus producing a 3D model with better quality and accuracy. Figure 6 shows the 3D model result of the Canseleri building at UiTM Shah Alam.
The front view had no gap or missing covered area during the image acquisition by using the DJI Phantom 3. However, there were a few spoils at the rear side of the model due to the lack of coverage at the narrow area. This was one of the limitations of using UAV, as a photo of the Canseleri building’s back alley could not be captured perfectly. There was a gap on the top of the building, as shown in Figure 7, due to the lack of camera coverage following magnetic and electric disturbances, causing the drone to lose signal while flying at the top of the Canseleri building.
The accuracy assessment was done by comparing the measurement of the 3D model with the direct measurement. The differences between the measured measurements with actual measurements could be compared. The actual measurements were collected using direct measurement, as previously shown in Plate 2. The measured measurements were collected from the software. The minimum number of samples needed to perform an analysis based on a normal distribution requirement is 30. As such, there were 30 samples analysed in this study. The normal distribution with 30 samples provided reliable results to estimate the accuracy of the reconstructed 3D building model. The number 30 came from the examination of the χ2 distribution, where 30 samples were normally needed to achieve short confidence bounds on the variance estimate. All samples were scattered evenly within the reconstructed 3D building model that covered all sides of the building. Several areas of the building were taken as the samples in order to compare the accuracy with the 3D model as shown in Figure 8.
Figure 9 shows the graph of the differences between the actual and measured values. It was found that the errors for all samples were less than 4 cm. There were a few spikes in Figure 9 because the selected sample had a large error. The large error might be contributed by the imperfect image matching during the build dense cloud stage and build mesh process. This was because the sample was hard to match due to the uncertainty of the geometry of the building. Parts of the building surfaces were covered with mirror and reflective objects. Therefore, image matching was difficult to perform on shiny or bright surfaces. The accuracy achieved for the reconstructed model was within 0.015 m. Therefore, the accuracy could be increased if more images were captured to fill in some gaps that might appear during image processing.
Conclusions and future work
The CRP method was used to achieve the 3D model of the Canseleri building using UAV as a platform for image acquisition. An oblique method with a random position of the UAV waypoint was used to acquire the images of the building. The UAV sped up the image acquisition process. This study proved that the UAV could produce accurate 3D model with textures. The accuracy assessment was done by comparing the measurement of the 3D model of the Canseleri building with the direct measurement of the Canseleri building using a measuring tape. In this study, the generated 3D building model had an accuracy of about 0.015 m. This result showed the UAV could produce a 3D building model at the centimetre level. Therefore, UAV can be used in the architecture or construction field to develop 3D models of buildings. The architecture or construction field requires a detailed and accurate 3D building model for inspection, renovation, progress monitoring, and maintenance purposes. The limitations of UAV must also be considered, such as signal noise, obstacle and light exposure during image acquisition.
Abu Hanipah, A. and Tahar, K. (2018), “Development of the 3D dome model based on a terrestrial laser scanner”, International Journal of Building Pathology and Adaptation, Vol. 36 No. 2, pp. 122-136.
Brunier, G., Fleury, J., Anthony, E., Gardel, A. and Dussouillez, P. (2016), “Close-range airborne structure-from-motion photogrammetry for high-resolution beach morphometric surveys: examples from an embayed rotating beach”, Geomorphology, Vol. 261, pp. 76-88.
Cavegn, S., Haala, N., Nebiker, S., Rothermel, M. and Tutzauer, P. (2014), “Benchmarking high density image matching for oblique airborne imagery”, International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences – ISPRS Archives, Vol. 40 No. 3, pp. 45-52.
Counts, B., Averett, W. and Garstki, K. (2016), “A fragmented past: (re) constructing antiquity through 3D artefact modelling and customised structured light scanning at Athienou-Malloura”, Cyprus Antiquity, Vol. 90, pp. 206-218.
El Meouche, R., Hijazi, I., Poncet, P.A., Abunemeh, M. and Rezoug, M. (2016), “UAV photogrammetry implementation to enhance land surveying, comparisons and possibilities”, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W2, 2016, 11th 3D Geoinfo Conference, Athens, 20-21 October, available at: https://doi.org/10.5194/isprs-archives-XLII-2-W2-107-2016
Genesio, L., Matese, A., Toscano, P., Bianconi, R., Zaldei, A., Gioli, B., Primicerio, J., Di Gennaro, S., Vaccari, F. and Belli, C. (2015), “Intercomparison of UAV, aircraft and satellite remote sensing platforms for precision viticulture”, Remote Sensing, Vol. 7 No. 3, pp. 2971-2990.
Ginovart, J., Toldra, J., Costa, A. and Coll, S. (2014), “Close range photogrammetry and constructive characteristic of masonry gothic vaults”, Journal of Construction, pp. 47-55.
Guidi, G., Russo, M. and Angheleddu, D. (2014), “3D survey and virtual reconstruction of archeological sites”, Digital Applications in Archaeology and Cultural Heritage, Vol. 1 No. 2, pp. 55-69.
Heilman, P., Gillan, J., Nichols, M., Sankey, T., McClaran, M., McVay, J. and Swetnam, T. (2018), “Considerations for achieving cross-platform point cloud data fusion across different dryland ecosystem structural states”, Frontiers in Plant Science, p. 8.
Hoang, N. (2018), “Detection of surface crack in building structures using image processing technique with an improved Otsu method for image thresholding”, Advances in Civil Engineering, pp. 1-10.
Mesas-Carrascosa, F., Agüera-Vega, F., Martínez-Carricondo, P., Pérez-Porras, F., Carvajal-Ramírez, F. and García-Ferrer, A. (2018), “Assessment of UAV-photogrammetric mapping accuracy based on variation of ground control points”, International Journal of Applied Earth Observation and Geoinformation, Vol. 72, pp. 1-10.
Mölg, N. and Bolch, T. (2017), “Structure-from-motion using historical aerial images to analyse changes in glacier surface elevation”, Remote Sensing, Vol. 9 No. 10.
Nex, F. and Remondino, F. (2014), “UAV for 3D mapping applications: a review”, Applied Geomatics, Vol. 6 No. 1, pp. 1-15.
Peppa, M., Mills, J., Moore, P., Miller, P. and Chambers, J. (2016), “Accuracy assessment of a UAV-based landslide monitoring system”, International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences – ISPRS Archives, Vol. XLI-B5, XXIII ISPRS Congress, 12-19 July, Prague, pp. 895-902, available at: https://doi.org/10.5194/isprsarchives-XLI-B5-895-2016
Perez, A. and Robleda, G. (2015), “3D virtualization by close range photogrammetry indoor gothic”, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, pp. 201-206.
Remondino, F. (2014), “Heritage recording and 3D modeling with photogrammetry and 3D scanning”, Remote Sensing, Vol. 2011 No. 3, pp. 1104-1138, doi: 10.3390/rs3061104.
Singh, S.S., Jain, K. and Mandla, V.R. (2013), “Virtual 3D campus modeling by using close range photogrammetry”, American Journal of Civil Engineering and Architecture, Vol. 1 No. 6, pp. 200-205.
Stöcker, C., Eltner, A. and Karrasch, P. (2015), Measuring Gullies by Synergetic Application of UAV and Close Range Photogrammetry – A Case Study from Andalusia, Catena, pp. 1-11, available at: https://pubag.nal.usda.gov/catalog/5424858
Vacca, G., Dess, A. and Sacco, A. (2017), “The use of nadir and oblique UAV images for building knowledge”, ISPRS International Journal of Geo-Information, Vol. 6 No. 393, pp. 1-26.
Ahmad Azmil, M., Tahar, K., Yusof, A., Nasro Ali, M., Mustafa, N. and Mahmon, N. (2016), “Geographical information system (GIS) map for fire and rescue application”, Jurnal Teknologi, Vol. 78 Nos 5-9, pp. 73-78.
The authors would like to thank Faculty of Architecture, Planning, and Surveying Universiti Teknologi MARA (UiTM), Research Management Institute (RMi) and Ministry of Higher Education (MOHE) are greatly acknowledged for providing the fund BESTARI 600-IRMI/MyRA 5/3/BESTARI (001/2017) and also the people who were directly or indirectly involved in this research to enable it to be carried out.