a pipeline for building generation
Diploma thesis - Erasmus Mundus Coperniucs Master in Digital Earth Program
Author: E.M.R.D Ekanayaka, richierde@gmail.com
Supervisors: RNDr. Stanislav Popelka, Associate Professor. Dirk Tiede
The study demonstrates the efficacy of a combined approach involving Esri CityEngine and ArcGIS Pro in establishing a systematic, user-friendly pipeline for generating LOD2 (Level of Detail 2) models with minimal expertise in the GIS domain. The pipeline automates the calculation of parameters such as eave height, ridge height, and various roof types, producing a feature layer usable within CityEngine for automatic LOD2 model generation.
The results shows the successful creation of LOD2 models for study areas while facilitating a streamlined pipeline that conceals workflow complexities from users. The generated pipeline offers comprehensive guidance for users, spanning from data collection to final accuracy assessment. The generated CGA script support modelling 15 roof forms that are most common in Czechia.
This section consist of important links such as map poster, tutorial videos presenting workflow process and a live ArcGIS scene
Urban settings bring unique problems and possibilities for efficient urban planning. However, traditional two-dimensional representations frequently fail to capture the complexities of three-dimensional systems, reducing the accuracy needed for informed decision-making.
Although 3D modelling softwares allow making LOD2 models it is often take a lot of time and manhours to adopt this process for entire cities. This limitation not only delays project timelines and escalates costs but also hampers scalability.
Esri CityEngine provides a potential solution for creating reusable CGA scripts. This research will introduce a systematic pipeline so that users can easily follow the procedure and generate LOD2 models of cities with minimal manual work. The results of the work will enrich the multidisciplinary fields of Geographic Information Systems (GIS).
Esri CityEngine provides a potential solution for creating reusable CGA scripts. This research will introduce a systematic pipeline so that users can easily follow the procedure and generate LOD2 models of cities with minimal manual work. The results of the work will enrich the multidisciplinary fields of Geographic Information Systems (GIS).
The following images exemplify some of the LOD2 models produced through our automated workflow. On the left side, you'll find representations with solid textures. Conversely, the right side images showcases models with realistic textures, offering a more immersive and lifelike experience.
These models are derived from comprehensive data analysis and processing, covering three distinct study areas.
The above figure depicts the RMSE of each research region for each ridge height estimation method. Overall, calculating ridge height using the median yields the least inaccuracy for residential and city centers. However, in rural areas, determining ridge height using the 90th percentile results in a lower RMSE, around 1 m less than the median technique.
The chart above shows that in all cases, the city center study area accounted for the highest RMSE. Aside from that, it is worth noting that in almost all circumstances, using the median to compute eave height is more consistent and accurate, with the exception of city centers, where the third percentile's RMSE is somewhat lower than the median.
Integer eu ante ornare amet commetus vestibulum blandit integer in curae ac faucibus integer non. Adipiscing cubilia elementum integer. Integer eu ante ornare amet commetus.
The model accurately captures the prevalence of flat rooftops, with a high PA of 0.89 and a good UA of 0.87. This shows that the model correctly identifies flat roofs when they are there and makes accurate predictions. Similarly, for gable roofs, the model shows a high PA of roughly 0.87 and a strong UA of about 0.96. This implies that the categorization model accurately recognizes and predicts Gable roofs.
The majority of roofs in this area are classified as flat, which shows a high producer's accuracy (PA) of around 0.89, demonstrating that the classification model accurately recognizes flat roofs when they exist. However, it is worth noting that the user's accuracy (UA) for flat roofs is slightly lower, at roughly 0.91, implying that other roof types may be misclassified as flat. While the PA for gable roofs is reasonable at 0.80, the UA is significantly lower, at around 0.71. This means that, while the model correctly recognizes many Gable roofs, it is more likely to misidentify other roof types as Gable.
The model's performance in recognizing flat roofs is relatively poor, with a PA of about 0.54 and a UA of roughly 0.73. This shows that other roof types are more commonly misclassified as Flat in this location. The model has a high PA of roughly 0.81 for Gable roofs, indicating that it can accurately identify them when they are present. However, the UA is slightly lower, around 0.76, indicating that other roof types were misclassified as Gable. Hip roofs have the lowest accuracy metrics in this region.
This study is a key milestone in the discipline of urban modelling and analysis since it provides a strong and scalable framework for creating LOD2 buildings in Olomouc and elsewhere. This project, which used advanced GIS technologies and computational tools, not only solved existing building generation difficulties but also opened the path for future developments in the sector.
The performance of the classification differs depending on the kind of roof and the location. The model does a good job of classifying some types of roofs, such flat and gable roofs, in the residential and rural sectors. However, it has trouble accurately classifying hip roofs, especially in the city center region. The model's ability to reliably categorize shed roofs is usually moderate across all locations, suggesting that there is space for development.
