True orthophotos produced automatically for large urban areas
In orthophoto projects of dense urban areas, true orthophotos are preferred over traditional orthophotos because they place the roofs of buildings in the correct horizontal position. However, there is still a widely held belief that the production of a true orthophoto is expensive and demanding. The authors investigated whether this was really the case based on a study of the production of an actual orthophoto from the municipality of Ljubljana, Slovenia.
Before producing a true orthophoto for the entire municipality of Ljubljana, Slovenia, the authors first tested various approaches in a smaller area in order to find the most economical workflow. One approach used a combination of a digital terrain model and a vector digital building model, and another approach used an automatically generated digital surface model. The resulting orthophotos were then compared to a traditional orthophoto in terms of aesthetic appearance and the manual work required. Based on the results, the team decided to use a true orthophoto based on an automatically generated digital surface model. The case study and the final project are presented below.
Differences between traditional and true orthophotos
An orthophoto is a photo or image corrected for projection distortions. It has a defined scale and can be used in the same way as a planimetric map. Orthophoto projects aim to provide a seamless orthophoto mosaic produced from simple rectified images. To produce an orthophoto, georeferenced aerial images and a digital reference surface model are needed. Traditional orthophotos have been produced around the world for over 30 years, providing an indispensable data layer in many GIS applications. In traditional orthophoto production, the reference surface is a digital elevation model. As a result, objects above the terrain (eg buildings, vegetation) are not shown in the correct horizontal position (Figure 1). In true orthophoto production, a digital building model is considered in built-up areas, or a digital surface model including vegetation cover is used. In true orthophoto production, algorithms have to solve two main problems: detecting hidden areas caused by objects above the terrain in the original image and preventing double mapping in these areas. To fill in the missing content in the masked areas, the images overlap at least 50% in both directions.
Study area and input data
The study area consisted of a densely built-up area in the center of Ljubljana (700m x 500m) and a suburban housing estate with residential houses (500m x 400m), thus covering two typical types of urbanization. Georeferenced aerial images with a ground sampling distance of 10 cm (GSD and 70% / 50% overlap, as well as a Lidar point cloud with a density of 18 points / m2 were the main input data. Both datasets were collected during the same aerial survey in April 2019. From a classified point cloud, a digital elevation model and a digital building model were produced (using TerraScan and TerraModeler of TerraSolid). A vector digital building model was first created automatically, but a lot of additional manual work was required to improve the model due to the complex building envelopes in the old city center. If the edges of the roofs are not precisely defined, double mapping occurs in the orthophoto (Figure 2).
A traditional orthophoto (in TerraPhoto) and two versions of a real orthophoto were generated from aerial images and a digital elevation model. In the first version, a combination of a digital terrain model and the digital vector building model described previously (in TerraPhoto) was used. The second version was produced in an almost fully automatic process in the nFrames SURE software. A digital surface model in the form of an irregular triangulated lattice was generated from a photogrammetric point cloud, produced with an image matching algorithm. Figure 3 clearly shows the advantages of a true orthophoto compared to a traditional orthophoto.
In addition, a visual comparison of the two versions of the real orthophoto revealed only small differences in the edges of the roof. In the version based on the automatically produced digital surface model, the edges of the roof are slightly toothed, which is a negligible flaw in the otherwise good overall aesthetic quality (Figure 4). One of the advantages of a true orthophoto produced from a digital surface model is that the trees are shown in the horizontally correct position (Figure 5).
Based on an estimate of the manual labor required to produce each type of orthophoto, the team concluded that the most labor-intensive approach is the production of a true orthophoto based on a combination of a digital terrain model and a digital vector building model. This requires about twice as much manual labor as the production of a traditional orthophoto. On the other hand, a true automatically produced orthophoto takes about 25% less time to produce than a traditional orthophoto. After considering all these aspects, the authors decided to apply the real automatic orthophoto production line in the operational project.
A real orthophoto for Ljubljana
Ljubljana, the capital of Slovenia, has approximately 290,000 inhabitants and covers an area of approximately 275 km2. Like any large city exposed to rapid changes in urbanization, the municipality needs up-to-date geodata for its decision-making. Nationally in Slovenia, a traditional orthophoto is available every three years in 25 cm GSD. However, this product does not meet the needs of the Municipality of Ljubljana. In 2020, the municipality therefore funded the creation of a real orthophoto mosaic, which was produced by Flycom Technologies. Aerial data collection was carried out in April 2020 with images overlapping 80% / 60% (and 80% / 80% downtown), with a GSD of 5cm. After performing the aerial triangulation of the images (in Match-AT by Trimble Inpho), a true orthophoto was produced (in SURE) based on a digital surface model (Figure 6). The estimated horizontal accuracy, calculated from the ground control points, was 0.04 m in both X and Y directions. As an example of data visualization, a photo-rendered 3D mesh of the city center was was created from nadir images (Figure 7). On the basis of these good results, the Municipality of Ljubljana has now decided to finance the annual production of a real orthophoto.
The success of this large-scale project shows that the automatic production of a real orthophoto is already fully operational in real life. The resulting true orthophoto mosaic is of good quality and requires much less manual labor than other approaches, saving a substantial portion of the final project costs. Needless to say, such an approach requires the right software, powerful computers, and an investment in staff training. However, these initial investments can be quickly recouped in subsequent projects. The authors would like to point out that a true orthophoto requires more aerial image overlap than a traditional orthophoto. However, this cost is only a small part of the final costs. Considering all the aspects discussed here, the authors conclude that there is no reason not to produce a true orthophoto almost fully automatically in urban areas.
Gharibi, H., Habib, A. (2018). True generation of orthophotos from aerial images and Lidar data: an update. Remote sensing, 10 (581), 1-28. DO I: https://doi.org/10.3390/rs10040581
Haggag, M., Zahran, M., Salah, M. (2018). Towards the automated generation of true orthoimages for urban areas. American Journal of Geographic Information System, 7 (2), 67-74. DOI: 10.5923 / j.ajgis.20180702.03
Nielsen, MO (2004). True generation of orthophotos. Memory. Lyngby: Technical University of Denmark, Computer Science and Mathematical Modeling. http://www.close-range.com/docs/True_Orthophoto_Generation.pdf
This case study was partly funded by core funding from the Slovenian Research Agency (No. P2-0406 Earth Observation and Geoinformatics). The authors would like to thank Flycom Technologies for data and technical resources as well as the Municipality of Ljubljana for permission to publish the results.