Web mapping is essential in many fields, including navigation, real estate, and tourism. With advances in mapping technology, 3D web mapping has been introduced by multiple geospatial platforms. Compared to 2D web mapping, 3D web mapping provides more details, better visual effects, and is crucial for applications like digital twins. In navigation, terrain is a fundamental element in 3D mapping. However, the vast amount of data from 3D models requires the use of multiple Levels of Detail (LoDs) to reduce data consumption and request time on the client side. Without further processing, different LoD models will have gaps at their boundaries. Such gaps can cause visual inconsistency and artifacts. Locking boundaries is a commonly used method to solve the gaps in different LoDs problems, that is to maintain the vertices on the boundaries during simplification. While this the method causes inefficiency in simplification. Nanite (Nan) proposed a method using triangle clustering to address issues with Level-of-detail (LoD) model gaps, inefficiencies in cluster simplifications, and visual inconsistencies caused by locked boundaries. This method can be applied to large-scale models, such as terrain data. However, this method was only applied to local data use cases and not to web mapping. Therefore, this thesis investigates adapting the triangle clustering method for web mapping. Considering the fact that the Nanite triangle clustering method performs well on multi-LoD models (Nan), we believe this method could also be migrated to the area of large-scale terrain. We proposed a method to construct multi-level LoD terrain for navigation based on Nanite method. In our approach, large-scale terrain data is distributed using web tiles for navigation purposes.
Keywords— WebGIS; Web mapping; 3D map; Multi-LoD

The accuracy and comprehensiveness of 3D city models have become increasingly important for applications like monitoring, sustainability evaluation, disaster management, and urban planning. However, creating accurate and complete 3D models is challenging. Traditional methods, such as photogrammetry and Light Detection and Ranging (LiDAR), often face issues like time-consuming processes and data gaps caused by occlusion. Additionally, these methods typically produce discrete models that fail to capture all the information from the original objects, limiting the ability to reconstruct small structures.

In this thesis, we explore the potential and characteristics of enhancing 3D models for urban areas using implicit neural representation. This method offers generalizability, ensures no void areas, and provides more detailed information when using multi-modal inputs. The datasets used for model training include Actueel Hoogtebestand Nederland 3 (AHN3), 2019 Luchtfoto Beeldmateriaal, 3D Basisregistratie Adressen en Gebouwen (BAG), and Basisregistratie Grootschalige Topografie (BGT). The city center of Eindhoven is used for training and testing, and the city center of Rotterdam serves as a test dataset.

The method involves learning location-dependent latent codes from raw point cloud and orthophoto, and adjusting the probability of space occupancy based on point clouds sampled from the 3D city model. A decoder is used to calculate the probability of existence for any point in the 3D space from the continuous field. Proper sampling allows us to derive a continuous Digital Surface Model (DSM) with unlimited resolution from the neural network.

The results show a high degree of accuracy: the generated DSM for the training area in Eindhoven demonstrated a median absolute error of 0.484 m overall. Accuracy for building areas was recorded at 0.798 m, and for terrain, it was 0.302 m. The model also displayed robust generalization capabilities, with an accuracy of 0.336 m in the Eindhoven test area and 0.23 meters in Rotterdam. Additionally, the model's ability to fill voids was confirmed through both visual inspection and quantitative evaluations. This research underscores the potential of implicit neural representation for generating detailed DSMs and effectively filling no-data voids.

Generative AI is developing rapidly and has shown great potential in generating and editing images with text prompts. It has achieved partial success in the challenging 3D model edits of common objects. However, there is a lack of attention in the building domain, which already faces certain limits in the 2D space. 3D building models automatic edit has wide application potential, especially for renovation, concept comparison, and large-scale scenes. Therefore, the thesis aims to edit the building models effectively through texture creation and geometric changes with generative AI by evaluating the existing 3D edit pipelines and putting forward practical solutions based on them.

The thesis first examines the current methods. Latent-paint, Text2Tex and X-Mesh are chosen as the representative existing explicit representation (mesh) based pipelines for their relatively satisfying performance in common objects and the different ideas contained. Modifications are made based on them to improve output quality from the aspects of the control module and edit module. For the first aspect, attempts include adding the view specification text prompt to the original X-Mesh and using the image as the control for X-Mesh. Image control X-Mesh is also further experimented with placing higher attention on the input image view and editing the key semantic parts of the building separately. For the latter aspect, modifications include freezing sampled vertices geometry and combining X-Mesh and Text2Tex.

Results show that Latent-Paint mainly creates texture with only major colour and lacks details. Text2Tex generally generates fine textures for inputs with richer depth information. X-Mesh creates textures and edits geometry jointly. It can improve the fidelity of the input mesh to some extent but suffers from noise problems. Image control X-Mesh generates more realistic results and outperforms the text control pipeline. Combining Text2Tex with X-Mesh takes advantage of Text2Tex, which can generate smoother and more realistic textures compared to X-Mesh in some cases. The combination is especially recommended for low-fidelity mesh of small or medium size. The user study also shows that the combination of X-Mesh and Text2Tex generates the most favoured results while image control X-Mesh ranks second. There are still limitations in geometric deformation scope, fidelity, generalizability and computational efficiency in the proposed 3D edit pipelines. The thesis succeeds in proving the application potential of 3D edit pipelines in the building domain and obtaining high-quality results by modifying existing methods.

Road network models are essential for modern urban planning, transportation analysis, and digital map reconstruction. This thesis addresses the development of a comprehensive digital road network model that advances from linear representation to a lane-level and areal polygonal representation, based on OpenStreetMap (OSM) data. While OSM provides rich road network data, challenges remain in reconstructing high-resolution, lane-level networks that are both geometrically precise and topologically consistent.

The objective of this research is to create a reffned digital road model that enhances the geometric accuracy and topological integrity of road networks by transitioning from road-level centerlines to lane-level polygons and full areal representations. This model is designed to capture essential aspects and semantic information such as traffic modes (motorized and cycling), intersection complexity, and road connectivity.

To achieve these objectives, the lane network generation methodology integrates OSM data with graph theory and civil engineering principles, focusing on lane topology and connectivity. The approach enhances linear geometry into areal geometry by combining a buffer-based method for road ribbons with a node-based method for intersection geometry. This process generates multiple intermediate results, including a road-level data model, a traffic-level data model, and grouped strokes representing global and local adjacency as graph-like models. Additionally, lane-level networks, lane geometry, and areal representations of roads and intersections are developed. Key innovations include a generalized method for lane centerline and polygon generation, ensuring road-lane correspondence and consistency between low-LoD linear models and high-LoD areal models.

The final product is a digital road network model that accommodates variations in road shapes, realistic intersection geometry, and detailed trafffc lane information, enabling highly detailed urban simulations and transportation analyses. While challenges remain regarding lane alignment, complex intersections, and the lack of detailed trafffc signal data, this research successfully bridges the gap between OSM’s raw linear road centerline data and the lane-level areal representations.

Geometric algorithms are usually designed based on the assumption of using arbitraryprecision representations. However, in modern computer systems floating-point arithmetic and finite-precision approximation are often used as implementing exact computations would require large computational resources and would be rather slow in the applications. Snap rounding (SR) is a widely used technique to transform representations of infinite precision into fixed-precision formats. It slightly modifies the original geometry of the input in order
to obtain a better organized geometric object arrangement and avoid the issues caused by applying floating-point arithmetic in geometric algorithms. The existing implementations of SR primarily focus on processing sets of line segments, while in practice polygons are of great significance as well yet has not received adequate attention so far. In this thesis, a new method is proposed to perform SR on 2-dimensional polygons. Firstly the input polygons are embedded into a triangulation. Subsequently, the triangulation is tagged with the ID information of polygons and the polygon boundaries are extracted and
stored in a container. The boundaries and the triangulation are then dynamically processed according to the identification of close polygonal vertices and close vertex and boundaries (under a predefined tolerance). After having snapped all the close elements the rounded polygons will be reconstructed from the processed boundaries. The testing results show the proposed method is capable of eliminating small gaps between elements (i.e. vertices and
boundaries) and is adaptable to various inputs. The topological and geometrical properties of polygons are preserved as much as possible. Finally, an overview of the limitations is provided, along with potential directions for future research.

With the development of technologies and people’s raising standards for both indoor and outdoor environments, buildings have expanded their functionalities of merely being human’s everyday sheltering place. Common modern building applications include but are not limited to Building Energy Modelling (BEM), solar potential estimation, wind simulations, shadow analysis, noise propagation, and digital building permit checking. Among all
the common building applications, many of them only need to use the building’s envelope (A building’s envelope is composed of all building elements that are exposed to the outdoor environment). It is apparent that building envelope extraction can be beneficial to these applications. Currently, building envelopes are mainly constructed from point cloud data or satellite image data. With the increased usage of Building Information Model (BIM) models within the Architecture, Engineering and Construction (AEC) industry, more attempts have been done on extracting building envelopes from BIM models as well. In addition, the BIM-based building envelope tool can be used throughout the building’s life cycle. However, previously developed BIM-based building envelope extraction tools all have different types of flaws. The main problem with the existing methods is separating the building envelope elements is difficult and time-consuming. Considering the benefits extracting building envelopes from BIM models can bring and the limitations of the developed methods, it is beneficial to develop a different BIM-based building envelope extraction approach that can produce high-quality building envelopes efficiently. Therefore, in this study, to avoid the difficult problem of separating building interiors and building exteriors, we first extract the point cloud from BIM models and then extract this point cloud’s exterior boundary using the 3D alpha shape algorithm, and therefore obtain the building envelope. The quality of the extracted building envelopes is evaluated by their geometric accuracy, simplicity, and time efficiency. Among them, the geometric accuracy is evaluated by the deviations between the extracted surfaces and the original surfaces. The simplicity is measured by the number of vertices and faces, and the time efficiency is measured by the building envelope extraction speed. The results show that despite the developed method has some limitations, it can extract building envelopes with high geometric accuracy from various types of small-scale BIM models. The main limitations include its inability to process big BIM models, the extracted building envelope merely contains its geometry without the enrichment of topological and semantic information. For all the tested BIM models, the average geometric accuracy is within 2.00cm. Walls and roofs are very accurately extracted, with errors of less than 0.01cm. Windows, doors, and other types of small-scale objects are extracted with a bigger error of
0.1m to 0.3m. The extracted building envelope also simplifies the input models significantly regarding the number of vertices and the number of faces. Lastly, for small-scale models, the developed building envelope extraction tool is able to process them within 150s.

The reconstruction of 3D city models has garnered significant interest in recent years. However, the majority of existing reconstruction methods primarily focus on LOD2 models, while LOD3 model reconstruction often relies on manual labor, and the primary data sources are street view images. This research aims to advance this field by reconstructing LOD3 models through the addition of windows and doors to existing LOD2 models, thereby maximizing the utility of available 3D building models, as well as the accurate addition of windows and doors. This research innovatively utilizes aerial oblique images as the data source for extracting building openings and employs 3D BAG LOD2.2 models as the basic 3D building structures. The 3D facades are projected onto the 2D aerial image space using perspective projection and registration is employed on the projection facade and oblique aerial images. Subsequently, Mask R-CNN is employed to detect and extract the building openings from these projections. Following the extraction, the layout of the openings within the same facade is optimized in terms of both size and position. Lastly, the relative positions of the openings on the facade images are combined with the 3D coordinates of the corresponding facade to calculate the positions of the openings in 3D space. This information is then integrated into the LOD3 model, resulting in a more detailed and accurate representation of the buildings.
This approach successfully reconstructs the final LOD3 model in CityJSON format, which passes the val3dity validation. By effectively utilizing existing 3D building models, this approach conserves a considerable amount of computational resources required for reconstruction. The simplicity and high level of automation of this approach make it a promising solution for reconstructing large-scale LOD3 buildings, leading to more accurate and detailed large 3D urban models.

Modern navigation heavily relies on Global Navigation Satellite Systems (GNSS) and digitized road network databases, but faces limitations in GNSS-denied areas and complex 2D road netowrks. This project addresses these challenges by developing a methodology to create and store a comprehensive 3D road and terrain dataset for enhanced navigation. In collaboration with TomTom, a company that aims to fulfill software requirements, making significant advancements in geolocation technology and societal contributions. The main research question of the project is: ”How can we create a 3D map of roads using information about the center of the road and elevation data?”. The approach to answer this question involves extracting 2D road polygons from centerline data based on width of the roads, the direction and the amount of lanes of them. These 2D polygons undergo enrichment with elevation data, with techniques like filtering, segmentation, and primitive extraction ensuring alignment with the digital terrain model. The methodology encompasses data acquistion, creation of polygons using the centerlines dataset, 2D-to-3D polygon conversion, elevation integration and data storage in CityJSON format.

Street view imagery (SVI) is one of the largest (growing) resources in urban analytics. A global close-up of the urban environment, if you will, which is rich in (untapped) information such as floor count. Floor count is useful in many applications, from improving energy consumption calculations to creation of 3D city models without elevation data. So far, efforts to extract floor count from SVI are mainly approached as a classification problem with the use of convolutional neural networks (CNNs). Limitations of this approach include the need of large (manually annotated) datasets, and uncertainty how these models learn to count storeys. Therefore, we aim to develop a method that can be trained on available datasets and determine floor count in a more explainable manner.

In order to make the floor count determination method more transparent, we mimic the row-wise counting of storeys as humans do: by vertically parsing a column of windows (and occasional door). Façade parsing is a common computer vision task that we can solve with deep learning. In this work, we employ the Mask R-CNN framework, that is trained on publicly available datasets, for the detection and segmentation of windows and doors. Then, the vertical distribution of detected / segmented windows and doors is estimated by computing the kernel density estimation function. The floor count is extracted by finding the number of maxima in the function, as the maxima represent the dense areas of windows and doors on a horizontal axis (i.e. storeys). To improve the results, an automatic image rectification is added as pre-processing step that enforces the regularity and repetitive occurrence of windows and doors. The full pipeline thus consists of three stages: 1) automatic image rectification, 2) window and door detection/ segmentation with Mask RCNN, 3) floor count estimation via maxima finding on the kernel density estimation (KDE) function. In addition, a small "wild" dataset was created that contains a higher variability in floor count, image quality and architectural styles, which better reflect real world SVI than existing façade datasets.

The floor count performance of the full pipeline was evaluated on the Amsterdam Facade (subset), ECP, TRIMS and "wild SVI" datasets. Since floor count annotations were missing, these are manually added. For detection-based data, the best results are an accuracy of 83% and a mean absolute error (MAE) of 0.17. For normalised segmentation-based data, the best results are an accuracy of 80% and a MAE of 0.20. Considering the method is still at its infancy, the results are promising. With further improvements in the pipeline and addition of automatic façade acquisition, the approach can contribute in large scale extraction of floor count information from SVI. To encourage further development, the pipeline prototype, dataset and floor count annotations are open source and will be released on https://github.com/Dobberzoon/Facade2Floorcount.

To keep the Panama Canal operational year round, the water level in the canal is maintained above the minimum depth. As water is passed through the locks and into the oceans on either side, the canals main reservoir looses fresh water, and this water is then replaced by the canals secondary reservoir which is Lake Alajuela. The Lake Alajuela reservoir is supplied of water by two large rivers, bringing in a sediment flux and decreasing the reservoirs capacity over time. Sedimentation sets a time limit to the use of the reservoir, and understanding this process and predicting the future morphological changes to occur in the reservoir are key to ensure continuation of operations in the Panama Canal. To predict the quantities of local sedimentation in the reservoir, a machine learning model is trained with features extracted from bathymetric models covering various years. These features are computed values that correlate with morphological processes. As such, the adaptiveness of a machine learning model along with the ability to extract morphological features from any elevation model provides the possibility to predict future sedimentation where other methods with empirical and numerical methods do not suffice due to a lack of data and restrictive parameters. Three types of machine learning algorithms and 21 features were initially tested for the purpose of the morphological modelling, during which the SVR model was the most successful. The testing was done on the R´ıo Chagres basin as well as the R´ıo Pequen´ı basin, both located in Lake Alajuela. Extensive hyperparameter tests were done to optimize and further test the performance of the SVR. Depending on the study area and the scale of the morphological behavior occurring, slightly different sets of features were most effective. Nevertheless, in both study areas tested the runoff model has proven to be a key factor for predicting the sedimentation and achieved a 70 to 80% accuracy for predicting zones of low or high sedimentation. The same model trained with the data from 1997 to 2018 was then used to predict 6 years into the future where more dynamic morphological behavior will occur according to the model. A downstream moving sediment front in the R´ıo Chagres basin is recognized and predicted to have moved 500 meters between the years 2018 and 2024. Such a sediment front travelling too far into the reservoir could have catastrophic consequences for all operations of the Panama Canal, making the awareness and anticipation of its progression highly important.

Automated reconstruction of detailed semantic 3D city models is challenging due to the need for high-resolution (HR) and large-scale input datasets, the ambiguous definition of the ensuing model, the intricacy of the processing pipeline, and its costs. Furthermore, existing methods mainly focus on geometry rather than semantics. Detailed semantic models may include roof installations whose size and function vary: dormers, windows, chimneys, etc. All elements visible on the roof from an aerial view are called ‘‘superstructures”. Deep Learning techniques can facilitate their modelization. This work inscribes itself in a project developed at the Technical University of Munich. The existing pipeline employs a convolutional neural network (CNN) on aerial images segmenting roof superstructures. These results can then be vectorized, extruded in 3D with their semantic description, and added to a simple 3D model.

This thesis demonstrates that building height data fused to a CNN on RGB aerial images improves the semantic segmentation of roof superstructures for classes with relief. Fusion of absolute and relative height data with different interpolation methods applied to LiDAR point cloud data is achieved through a fusion network from the state-of-the-art (FuseNet). First, experiments prove that detection accuracies increase by 11% on average for dormers and 12% for chimneys compared to U-Net output on the same dataset. Best performance is reached with the fusion of absolute height (rather than normalized) and IDW or NN interpolation technique (rather than none). However, although superstructure types are better recognized, their boundaries are fuzzier due to data input mismatches, and more background pixels are classified. Secondly, the predictions and modelization of both a Bavarian and Dutch test set prove the technique scalability. However, a training set annotated for Bavaria and applied to a test set in the Netherlands yields inaccurate results due to local architectural typologies and different input data characteristics.

The flexibility of the Industry Foundation Classes (IFC) format makes it challenging to blindly rely on the data it stores. This is because, to allow its flexibility, the stored data in an IFC file can be missing, incorrect, redundant, or stored in vague ways while the file can still be considered format compliant. This missing reliability is an issue for a format that is so central to the Architecture, Engineering and Construction (AEC) industry. This also severely hampers the chances for automated processes to be applied in the AEC industry. Chances that could reduce the time and cost spend on a construction while enlarging its quality. Prior done research on this subject is sparse. The small subset of this research that covers the ”repair” or reliable extraction of IFC data primarily focuses on idealized situations that do not often occur in practice. Due to the combination of these factors, this master thesis attempts to discover if it is possible to extract reliable data from the unreliable IFC files with an automated method. Because of the large flexibility and broadness of the IFC format, the scope of the thesis had to be limited. This was done by focusing on the extraction/reconstruction of storeys (and their elevations), rooms, and apartments. For these three subjects not only extraction/reconstruction methods have been created, but they have also been implemented in an open source software tool1. To develop reliable processes, the processes have to be based upon reliable data. From the data available in the IFC format, the tangible geometry is considered to be the most reliable. However, not all the other data can be discarded. Only the data that is implicitly stored in the tangible geometry can be ignored without losing information. In practice this means that most, if not all, topologic data and a subset of the semantic data can be ignored. It was found that for the extraction/reconstruction of storeys (and their elevations), a z-value extraction based on created IfcSlab/IfcRoof groups was fairly effective. These groups were made based on their size, height, and location. This process does show a lowered reliability in buildings with complex storey structures. It was found that for the extraction/reconstruction of rooms, a voxelization approach functioned very well. The boolean refinement that followed, which took the shape from a rough voxelization to the precise room shape, showed some reliability issues. Finally, it was found that for the extraction/reconstruction of apartments that rules applied to a created space syntax graph showed promise. The rules that were used were simple, resulting in lower reliability in structures that had more complex room relationships. In conclusion, this research has been unable to give a definite answer to the issues at hand. However, all three the subjects that have been covered showed promising results and can be considered steps towards a solution.

In the field of Geographical Information Science (GIS), geodata transformation and analysis tools often take the shape of software libraries written in system level programming languages.
However, the societal impact of these tools is often limited, as most end users only access these libraries via indirect means: Through bindings in other languages, through plugin in applications, or both.
Additionally, the tools end users end up with are often not composable, and may contain other hurdles like installation or configuration.

The goal of this study is to make core GIS libraries more directly available and composable to end users.
This study presents and prototypes a novel method, centered around a visual programming language to host the functionalities of GIS libraries from within an application, and in a composable manner.
Additionally, the visual language is used to connect these libraries to a user-definable Graphical User Interface (GUI).
This prototype is implemented as a static web application, so that these libraries are directly accessible to end users without installation.
GIS libraries are compiled to WebAssembly, making the library usable in any language, including
this web based visual language by using a ’no boilerplate’ plugin system.
Finally, both scalability to handle sizable datasets, and a rich GUI (3D viewers, file inputs, sliders), are primary design considerations and assessment criteria.

The results show that this specific web based VPL appears to be a feasible method for providing direct access to some native GIS libraries, and does offer a unique set of features not found in comparable visual languages.
The significance of this method, compared to other web based geometry VPLs, lies in the fact that it offers a lenient plugin system, in combination with a range of different GUI nodes, certain "dataflow VPL" properties, and a proposed zero cost abstraction runtime.
All of these features combined lead to a VPL which is able to directly connect GUI components with native GIS libraries, all while remaining scalable in principle.

On a practical level, more work remains to proof this feasibility.
The methodology developed by this study is only theoretically accessible and composable, based on achieved features.
User testing is required to confirm if this method indeed improves workflows, and actually saves time and energy of developers and end users.
Moreover, the prototypical software implementation used is limited and not production ready. Both the fact that the ’no boilerplate’ plugin system cannot be used with C / C++ GIS libraries, and that backend execution is not possible yet, must be improved upon in future work.
Despite this, visual programing, distribution using WebAssembly, and Rust based geocomputation, all proved to be valuable directions of future GIS research.

The plastic pollution of aquatic environment is undoubtedly an emerging environmental risk, as it negatively affects ecosystems globally to a great extent. To prevent the plastic soup from growing even further, a Delft-based start-up Noria has developed plastic collectors, to remove plastic from rivers and canals before it reaches the ocean. In order for these devices to give maximum positive effect, they need to be installed in areas where plastic is more likely to accumulate - the plastic hotspots. Taking into consideration various natural attributes that affect the movement of the plastic waste in the water, such as wind direction, water flow, canal geometry, vegetation and man made structures in waterways; potential hotspots can be predicted in a model which would allow more efficient coordination of the cleaning process. Thus, this project aims to locate plastic accumulation zones in the city of Delft in a (semi-) automated manner using open spatial data analysed in GIS and a network simulation model.
The methodology developed in this project results in the visualisation of potential plastic hotspots where Noria’s collectors could be placed in order to remove and recycle the plastic. The potential hotspots suggested by the model were compared with ground truth data collected. The final result yielded only 20% accuracy and therefore did not meet the initial expectation. An evaluation of the shortcomings was made with suggestions for future research.

In recent decades, it is well known that transportation systems have a crucial role in the economic development and social prosperity of modern societies. While road width can be considered as one of the most important factors of the road environments, there is no clear definition of it. To better understand this, we need to reconsider the width of a road not as a single numerical value but as a complex concept that can be interpreted in different ways. Different road users may refer to different width values for the same road. Thus, while estimating road width seems to come with great benefits for different real-world cases, a fundamental rethinking of the purpose of this entire process is necessary.
Road safety management is an application whose overall process is strongly affected by the width of the road. While there are conflicting theories about the effect that road width can have on road safety, we may need to reconsider a few things before exploring their relation. In this thesis, we introduce a novel approach for estimating road width and linking it with roads in such a way that the overall process of road safety management application could benefit.
Moreover, even road width estimation is not a new topic most of the studies so far use LiDAR point cloud or satellite images as input. In our approach, vector data coming from open-sources are used. Different inputs come with different limitations. The advantages, as well as the drawbacks of the different inputs, will be discussed.
Finally, one of the main objectives of this project is to develop a methodology that will be generally applicable. Vector data that can be found in different datasets would be used. Some additional deliverables will result from our efforts to overcome some of the challenges that have arisen in achieving this goal. A methodology that standardizes road vector data and a methodology that identifies the location and the type of the different intersections will be developed

In many countries digital maps are created and provided by the national cadastres: Usually they consist of multiple polygons, each with an exact location and shape, describing which kind of surface can be found at the position of the polygon (e. g. building, street, vegetation). They must be accurate and well maintained, as they are used by companies or authorities for purposes like urban planning or demographic statistics. However, especially cities are in a constant change. Old buildings are torn down, new buildings are built, and complete streets and neighbourhoods are changed. Monitoring these changes is difficult and identifying and updating the virtual maps is still done mostly manually today. A method is developed to detect changes on the ground and identify changes for the virtual maps automatically using machine learning approaches. As input data the virtual map, their corresponding aerial images and point clouds from different years are needed. As a case study for this thesis, this method is developed and applied to the BGT, the Dutch virtual map with a resolution of 20cm. The research area is the city of Haarlem for 2017 and 2018. High resolution aerial images are used in combination with point clouds created by Photogrammetry. The output is again a digital map of the area where every polygon has a probability score of how likely its category (for example building, street, etc..) changed. This can support the manual updating process eminently, as a minor percentage of polygons (for which the algorithm was unsure) must be checked manually. The research question of this thesis is to check whether this change detection is feasible even for highly heterogeneous structures like cities. Many visual changes in the aerial images are happening that are not relevant for the virtual map. In the one year, a street can be full of colourful cars, in the other year, the street is empty and completely grey. Many scenes are easy for humans to distinguish but are challenging for an algorithm. The goal is to detect a high amount of true changes while keeping the number of false positives low to reduce the manual work as much as possible. To achieve good change detection and answer the research questions, the machine learning library of XGBoost is used. It provides a gradient boosting framework for many different environments, including Python. Many weak learners, each classifying a change only with a very low detection rate (for example minimum height of all points inside the polygon), are combined to get a strong learner. This learner should be able to classify polygons with a high accuracy into polygons that change and polygons that do not change. With this method it is possible to detect a high amount of changes. 80% of all changes can be found within a reasonable number of False positives. Especially for buildings almost all changes can be identified. It is furthermore possible to localize the changes in larger polygons. However, not all changes can be identified, so that this approach should be seen as an aid for manual change detection and not to replace it.

Augmented Reality (AR) is the technology that superimposes digital generated objects on the physical world. It has the potential to create new products and services,like visualizing future buildings or objects that will decorate a place, which creates new opportunities for applications in both the public and private sector. The architecture and construction fields are particularly interested in investigating this innovative technology to engage stakeholders, such as designers and engineers, in every stage of decision making and to minimize discrepancies between the original design of a building and the final outcome. Portability is one of the greatest advantages of Head-Mounted Display (HMD) AR technologies, but there are limitations regarding the amount of data which can be visualized using the computational power of the current generation of devices. Also, automated methods and approaches that can cope with the intricacy of the models have to be produced and tested to make the usage of augmented reality feasible.Within this thesis, a methodology is developed to isolate each storey of a Building Information Modeling (BIM) model and its exterior envelope. Firstly, the model was converted, from the Revit file format, to the open standard Industry Foundation Classes (IFC), which made the file human readable. Only semantic information was used to isolate each storey of the building, while for the extraction of the outer shell geometrical calculations were performed. This extraction took place by sending rays from one side of the model to the other and checking the intersection of the rays with the model.Afterwards, the storeys and the exterior were visualized through an AR device,the Hololens. The Unity platform, which provides many tools for holographic development, was used for the configuration of the scene where the final user will interact with the created models. Scripts in the C# programming language were developed to allow interaction between the user and the holograms. I created a simple and intuitive menu, consisting of 3D buttons to allow the user to visualize only the desired parts of the model. After having visualized the model, the user has the ability to scale and rotate the model using the corresponding buttons. In the same way, when the user stares at an element of the building, this element is high-lighted and by making the tap gesture, he/she can visualize metadata information about this element as text above the model. Finally, spatial perception functionality is provided by virtually placing the model on horizontal planes identified by the device.The proposed methodology was tested in a use case on a sample BIM model, and specifically of the Amsterdam Medical Center. The large size of the file and the high complexity of its geometry made the model a challenging test that made it possible to highlight the limitations and efficiency of the developed approach. Despite the positive results of the process, the accuracy is affected by the computational power of the current generation of hardware. Nevertheless, the clear perception of a construction coupled with the interactions capabilities provide an immersive experience which can actively involve the user with the visualization process