Adhisye Rahmawati
…Too cool to be true? Cooling demand based on the (enriched) semantic 3D city model of Rotterdam
The DigiTwins4PEDs project explores the role of Urban Digital Twins in advancing Positive Energy District (PED) initiatives. This thesis will contribute to the project by simulating cooling demand in buildings of selected study areas within the municipality of Rotterdam, e.g., in the district (Wijk) of Feijenoord. The objective of this research is to simulate the energy demand for cooling within the study areas using a CityGML-based 3D city model. The research methodology involves enriching the city model with energy-related data needed to estimate cooling demand. Algorithms will be developed to detect buildings with insufficient or inaccurate data, as these inaccuracies may cause discrepancies in cooling demand simulation results when comparing outcomes from different simulation methods. SimStadt and Nieman models will be used to simulate and compare cooling demand results. However, the Nieman model has limitations, such as the lack of 3D geometry integration, which leads to less reliable estimates. Furthermore, the simulations will consider the year 2050 to account for projected climate changes, which are expected to increase cooling demand due to rising temperatures. The simulation results will be analyzed, and a pipeline will be defined to store significant results into the 3DCityDB extended with the Energy ADE. This research also includes evaluating the accuracy of the simulated cooling demand and exploring effective methods for presenting these values.
Supervisors: Dr. Giorgio Agugiaro and Dr. Weixiao Gao
(company involved: Gementee of Rotterdam: Mr. Frerye Hechavarria)

Bart Manden
The effects of building model automatic reconstruction methods for CFD simulations
Computational Fluid Dynamics (CFD) is becoming increasingly important in addressing challenges in the urban environment, driven by the rapid urbanization and the impacts of climate change. While existing studies highlight the importance of CFD for urban analysis and planning, a crucial gap remains regarding the types of geometries used in CFD simulations. As the urban sector embraces digitalization, more data is available to create detailed city models for CFD applications. This thesis will investigate the effect of various levels of voxelization in urban model reconstruction on CFD outcomes. By exploring how different voxel resolutions influence model quality, this thesis aims to assess their impact on the accuracy and reliability of CFD simulations. Ultimately, the findings may contribute to establishing guidelines for effective modelling approaches in CFD applications.
Supervisors: Clara Garcia-Sanchez and Jasper van der Vaart

Citra Andinasari
Point Cloud for 3D Cadastral
The rapid growth of urban areas has led to the development of increasingly complex high-rise buildings. This growth necessitates a land administration system (LAS) capable of optimally storing and visualizing the legal status of these structures, ideally through a 3D cadastral system. Building Information Modeling (BIM) has been widely used in various studies to create 3D cadastral systems, demonstrating great potential for representing LAS. However, as not all buildings have BIM data available, it raises the question of how to address this limitation. Recent studies have used point clouds as the basis for creating digital twins, where object semantics can be derived through automatic point cloud segmentation. As point cloud data is more abundant than BIM data, this study aims to explore the extent to which point clouds can be used to visualize 3D cadastral information. To answer this research question, three sub-questions are proposed: 1. How can point cloud data be processed and semantically enriched to identify and represent cadastral features and boundaries? 2. How to integrate 3D point cloud into LADM (Land Administration Domain Model)? 3. Which web architecture is suitable for visualizing the resulting 3D cadastre?
Supervisors: Peter van Oosterom and Edward Verbree

Der Derian Auliyaa Bainus
Integration of Geospatial Point Clouds Across Island, Using Features as Ground Control Points Extracted From Aerial Image
This project focuses on integrating geospatial point clouds from separated islands into a unified 3D model, addressing challenges such as spatial separation, lack of overlap, and variability in data resolution and accuracy. Traditional alignment methods relying on overlapping areas are often ineffective in this context. Thus, the project explores using alternative ground control points (GCPs) extracted from aerial imagery to integrate point clouds. Various methods for extracting features as GCPs from aerial imagery and point clouds can impact integration accuracy. In aerial imagery, techniques like manual point picking, supervised classification of building roofs, and other feature-based methods help identify GCPs, while in point clouds, features like building roof corners or road markings are used for integration. Different combinations of these methods may yield varying accuracy levels, influencing integration results. The primary research question is: “How can geospatial point clouds from multiple, separated islands be integrated into a seamless 3D model using features as ground control points identified from aerial images?” This project aims to create a framework for integrating geospatial point clouds from separated islands by identifying suitable features as GCPs from aerial imagery, testing existing feature extraction methods, and evaluating how aerial image resolution and feature extraction choices affect model accuracy.
Supervisors: Daan van der Heide and Jantien Stoter

Dimitrios Mouzakidis
LADM-Based Digital Twin for Archaeological Site Information Registration
Dense urbanization has led to increasing demand and pressure for land development, resulting in the partition of 3D space into different owners sharing delimited property interests on, above, or below the land surface. Consequently,(cadastral) spatial units range from simple, but most common, 2D, to complex 3D collections of spaces worldwide, that are more difficult to handle in terms of surveying, storing in a database, maintaining, visualizing, etc. This thesis project focuses on creating a web-based 3D cadastral prototype to accurately register and visualize geospatial information related to archaeological sites and their surroundings, integrating both legal and physical spatial data using the Land Administration Domain Model (LADM). Thus, exploring point cloud data for representing 3D structures and highlighting the relationship between legal spaces (e.g. ownership and jurisdiction) and physical objects (e.g. archaeological sites) would ensure a holistic view of the spatial arrangement. This LADM-based approach will enhance the integration and interoperability of archaeological records with existing cadastral data, supporting comprehensive lifecycle management and ensuring these valuable cultural assets are preserved within a legally standardized framework. In this context, relevant 3D storage and visualization standards related to data format, grammar, and implementation as with APIs and Web Feature Services should be explored (proposed by ISO, OGC, and W3C).
Supervisors: Peter van Oosterom and Eftychia Kalogianni
(company involved: CGI NEDERLAND B.V (Robert Voûte, Vice President Consulting Geo-ICT))

Georgios Iliopoulos
(Semi-)automatic modeling of indoor building 3D models for daylight simulation with Lidar-enabled mobile devices
Practitioners, such as architects and building technology specialists, need buildings' plans or 3D models for physics-based decision-support tools when planning retrofit interventions. In some cases, these plans are outdated or unavailable. Generating them can be time-consuming, so technologies like LiDAR scanners are used. Apple has equipped the Pro and Pro Max versions of the iPhone with a LiDAR scanner, providing access to advanced technology at a lower cost than professional equipment. Additionally, Apple’s RoomPlan API can create a 3D floor plan of a room, capturing walls, windows, doors, flooring, and furniture types using the LiDAR scanner, device cameras, and machine learning models. Leveraging the API’s output, this thesis will examine its potential as a low-cost data source for 3D models in daylight simulations for indoor spaces. For accuracy, these models must be watertight, with walls, floors, and ceilings free of gaps, and explicitly modeled doors and windows. With available bounding-box data, furniture will also be included. Materials of all entities must also be modeled, as their reflection properties greatly impact simulation results. The accuracy of the generated models and their impact on daylight simulations will be assessed by comparing them with manually generated models representing the ground truth.
Supervisors: Ken Arroyo Ohori and Eleonora Brembilla

Haohua Gan
3D reconstruction of linear urban objects from aerial lidar point clouds
The 3D models play a crucial role in urban planning, providing essential information for infrastructure design, land-use assessment, and environmental analysis. Most current reconstructions of 3D models focus on volumetric objects such as buildings, yet detailed surface models of linear urban objects, such as power lines and electrical poles, enable urban planners to conduct more precise planning and decision-making processes. At an urban scale, aerial lidar point clouds serve as a valuable tool for capturing high-resolution 3D spatial data of the built environment, offering a foundation for generating these surface models. However, for large-scale point clouds, reconstructing individual instances of linear urban objects is challenging due to factors such as noise, occlusion and insufficient point density, which limits the detail and accuracy of the surface representation. This thesis aims to address these issues by developing improved reconstruction techniques tailored to linear urban objects, enhancing model accuracy and usability. The plan is to focus on power lines and electrical poles, follow the extraction-reconstruction-optimization structure, and discover a possible primitive-based or skeleton-based reconstruction method.
Supervisors: Hugo Ledoux and Weixiao Gao

Hidemichi Baba
Cloud Optimization Strategies for CityJSON in 3D Urban Modeling
Standardizing data formats for 3D city models is vital for semantically storing real-world information as permanent records. CityJSON, an OGC standard, and its variant, CityJSONSeq, simplify data utilization by software. However, the shift to cloud-native environments and growing demands for processing large datasets require more efficient solutions. While formats like PMTiles, FlatBuffers, and Cloud Optimized GeoTIFF address vector and raster data, options for 3D city models remain limited. This research explores optimized data formats for CityJSON tailored for cloud-native processing, focusing on FlatBuffers. Features such as spatial indexing, spatial sorting, and partial HTTP Range requests will be implemented. The study includes a review of performance-optimized formats, adaptation to enhance CityJSON, and benchmarking. Results will enable end-users to efficiently download 3D city models and allow developers to store large datasets in single files, simplify cloud architecture, and improve processing. This work aims to enhance the scalability and usability of 3D city models, advancing urban planning and smart city applications.
Supervisors: Hugo Ledoux and Ravi Peters

Hsin-Yu Cheng
Orthophoto-Driven LoD2 Building Reconstruction: Balancing Geometric Constraints and Deep Learning
This project focuses on developing an efficient and cost-effective framework for LoD2 building modeling using only orthophotos. Traditional methods often rely on costly height data, such as LiDAR or DSM, making them inaccessible in resource-constrained regions. By leveraging geometric constraints and deep learning, this study aims to accurately reconstruct rooftop topologies without height information. Geometric constraints filter noisy or incomplete line segments, while deep learning enhances classification and refines structures using RGB features. A Markov Random Field (MRF) model integrates these results for improved plane detection. To ensure robustness, DSM or LiDAR data will be used as validation tools to compare performance under scenarios with and without height information. The proposed framework reduces modeling costs, extends accessibility, and supports scalable applications in urban planning, disaster management, and digital heritage preservation.
Supervisors: Liangliang Nan and Weixiao Gao

Hyeji Joh
Generating Synthetic Distributions of Housing Units for Modelling Economic Segregation
Social simulations often overlook spatial frameworks, relying on oversimplified, abstract configurations, or real, context-dependent layouts. This research seeks a middle ground by developing a generator of synthetic distributions of housing units within Dutch urban layouts to study economic segregation, and to address the importance of spatial factors in modelling urban economic disparities. The approach involves analyzing geospatial characteristics (such as spatial patterns, density, land use, building types, and more) unique to Dutch cities, defining standardized, adjustable parameters, and integrating these into a spatial generator. The generator outputs a synthetic city environment where housing distributions reflect Dutch urban characteristics specified by the user-defined parameters, such as the number and type of households, based on the location in the city). The methodology supports urban planners, researchers, and policymakers in simulating housing distribution scenarios and studying their potential effects on economic segregation. A key challenge is determining whether to include socioeconomic dynamics in generating the synthetic city environment, as these may bias the model by influencing urban form and distribution. The anticipated outcome is a user-driven synthetic environment for simulating urban layouts and housing allocation to understand better the spatial dynamics of economic segregation, which can be used in simulations of economic segregation.
Supervisors: Hugo Ledoux and Clémentine Cottineau

Javier Martínez
Using thermal imagery to differentiate rock and soil types with deep learning techniques
A wide range of fields such as geology, construction and urban monitoring and planning rely on accurate knowledge of the ground for different applications and purposes. Emitted radiation by Earth's objects are recorded by thermal imagery, determining the temperature of objects in the Earth's surface. On the other hand, deep learning techniques can provide a more accurate and detailed detection and classification of the data. The thesis will explore the use of thermal imagery in combination with deep learning techniques to distinguish between different types of soils and rocks on Earth's surface. Thermal imagery captures heat signatures from materials, therefore proving to be a key source for geological identification. It will also leverage deep learning techniques to handle complex, non-linear relationships inside the datasets. Convolutional Neural Networks can be an effective resource for image classification to distinguish varying attributes, while additional Recurrent Neural Networks can analyse temporal variations for further details over time. The objective is to develop a model that can classify rock and soil types based on their thermal characteristics for varying conditions to ensure robustness over several environmental factors. To do so, the idea is to convert raw thermal data into structured datasets so that the trained models can work with (normalization, image clarity, thermal patterns). Image source: Soil temperature map (Peipei Lin, 2015).
Supervisors: Azarakhsh Rafiee and Remi Charton (PhD MsC Civil Engineering building)

Jessica Monahan
Designing Climate-Responsive Cities - A Morphology-Based Tool for Urban Wind Flow and Thermal Comfort Analysis
City designers face a challenge in predicting urban climate factors like air quality, ventilation, and heat stress. Urban form, including building density, layout, and orientation, affects local climates by shaping wind patterns, pollution dispersion, and heat retention. However, planners lack accessible tools to visualize and quantify these effects in specific city settings, as current studies often rely on idealized scenarios, creating a gap between practical needs and available tools. This thesis aims to bridge that gap by developing a morphology-based tool that allows urban designers to assess climate impacts for specific layouts. The project will focus on analyzing urban wind flows and thermal comfort using features of the UMEP tool. The goal is to create a framework for an online tool that allows designers to visualize current urban climate conditions in the Netherlands and assess the impact of proposed designs. The research follows three steps: first, building a data pipeline that automatically generates UMEP tool input data for specified locations. This pipeline will also allow designers to adjust building characteristics to evaluate climate impacts. Second, the tool’s outputs will be validated against established methods, such as CFD simulations. Finally, the tool will be tested on real-world cases to analyze how various urban morphologies influence local climates. This approach aims to make climate-sensitive design more accessible for planners and architects.
Supervisors: Clara Garcia-Sanchez and Hugo Ledoux

Lars Boertjes
Adaptive Mesh Refinement for 3D Reconstruction of Buildings Using the Segment Anything Model
This research focuses on improving advanced methodologies for generating 3D meshes from oblique photographs using Artificial Intelligence. By leveraging the Segment Anything deep learning network, the project aims to accurately segment buildings and possibly other identifiable features within oblique images, which will later be incorporated into the 3D generated mesh. Additionally, when creating segmented buildings within the 3D mesh, the goal is to allocate more resources to the more detail-intensive areas of the building during mesh generation. Existing end-to-end frameworks like Deep3D leverage Structure-from-Motion (SfM) and an adaptive Multi-View Stereo (MVS) model to create a dense point cloud, from which a mesh is then extracted. However, the resulting mesh lacks semantics and is simply a collection of faces and vertices without any inherent meaning. Our objective is to investigate whether we can use the Segment Anything network to: 1) Create a mesh with semantic information, or at least segmented building objects. 2) At the building level, further segment the components and use this segmentation to determine which parts of the building require more detailed modeling. By doing so, we aim to create an adaptive point distribution (i.e., allocate more points from the dense point cloud) based on the geometric entropy of the segmented building. For example, a flat façade requires fewer resources than a more complex
Supervisors: Azarakhsh Rafiee and Ken Arroyo Ohori
(company involved: Annemieke Verbraeck)

Lars Cornelis Huizer
Point Cloud and Shadow Fusion: A Framework for Height Estimation of Difficult Objects and Shadow-Structure Association Using AI Vision Models
This thesis will explore a new methodology for estimating buildings heights and enhancing structure recognition in urban environments by leveraging shadow analysis in combination with 3D point cloud-based surface reconstruction and AI vision models. Conventionally, building height estimation often relies on either LIDAR scanning or high-resolution aerial imagery, where shadow length is used to infer height. A technique will be proposed that reconstructs surfaces from point clouds to generate 3D urban models, on which shadows are simulated and analysed to detect link specific shadows to their corresponding buildings more accurately. The end result of this combined method hopes to achieve more accurate height estimation for difficult objects such as antennas, than the sum of its parts would provide--that is, LIDAR and aerial photography.
Supervisors: Dr. Azarakhsh Rafiee and Ir. Edward Verbree
(company involved: Antea Group, exact contact person unclear)

Lotte de Niet
Reconstructing 3D apartment units from legal apartment drawings
To provide more insight into the ownership rights of apartment buildings and enable integrating the 3D apartment units in other 3D data sets like the 3DBAG, the Dutch Cadastre (Kadaster) is working on automating the conversion of analogue floor plan deeds into 3D models. Previous research by the Kadaster has focused on vectorizing the deeds, and identified the main further challenge as automating the process of georeferencing and scaling. My research will address the question: How can 2D vectorized ownership deeds be georeferenced and reconstructed into a 3D model? I propose using the vectorized deeds as input and research the reconstruction process with the following steps: Step 1. Georeferencing In order to register the 3D model to the correct location, it needs to be georeferenced. The focus of the research lies in the automating the process of finding accurate anchor points in the deeds, and assigning geographical coordinates to them. Step 2. 3D reconstruction from 2D polygons to 3D volumes The plans can then be lifted into 3D using the height and number of floors estimates from the 3DBAG. Solutions to various scenarios will be researched, such as multilevel apartments, split-level floors and floors that are shifted rather than aligned directly above one another. The primary goal is to create a simple visualization, but if time allows, aligning the models with BIM legal standards and alternative approaches to visualizing ownership could be explored.
Supervisors: Jantien Stoter and Amir Hakim
(company involved: Kadaster, Bhavya Kausika)

Marieke Margaretha van Arnhem
Temporal Change Detection and Visualization in Point Clouds
Point clouds are the raw data of the real world, change detection analysis between the point clouds enables analysts to detect, quantify and interpret spatial changes directly. This thesis aims to develop and explore methods for detecting significant differences between point clouds generated at different times. The focus is not on identifying the differing points, but on recognizing the actual objects that have been added or removed. Key considerations in this thesis include how to effectively visualize these differences: should the raw data or the reconstructed object be displayed when an added or removed object is shown? Additionally, the research will address how these changes should be incorporated into the dataset. While the algorithm is tested using the AHN datasets, it is designed to be adaptable to point clouds of varying densities.
Supervisors: Ir. Edward Verbree and Prof.dr.ir. Peter van Oosterom
(company involved: Geodelta, Annemieke Verbraeck (annemieke@geodelta.com))

Michalis Michalas
Super-Resolution and Domain Transfer for Enhanced Aerial Imagery
Aerial images are key for urban planning, environmental monitoring, and geographic analysis, yet limited resolution often reduces detail and constrains outputs like object detection and digital surface modeling.High-resolution aerial imagery incurs significant costs for advanced cameras and substantial storage, limiting its availability to certain seasons with only two captures per year—one at high resolution and one at low.This restricts detailed analyses across all seasons.This thesis explores using single-image super-resolution (SISR) to generate high-resolution details from low-resolution aerial images over the Netherlands, aiming to improve accuracy and visual quality. By evaluating deep learning approaches, the research seeks methods to transform 25 cm images into refined 8 cm outputs, balancing accuracy and detail. Domain adaptation techniques will be tested to enhance SISR performance on both synthetic and real aerial images, ensuring models work across data environments and seasons. Creating “synthetic” high-resolution images can enhance DSM rasters, TrueOrthos, and solar maps by providing clearer, low-noise inputs. Though these synthetic images may not match original 8 cm quality, they should offer a marked improvement over 25 cm images, bridging the gap between low and high resolutions and supporting Readar B.V.’s production pipeline with consistent, detail-rich datasets across the Netherlands.
Supervisors: Martijn Meijers and Azarakhsh Rafiee
(company involved: Sven Briels)

Michele Giampaolo
The Impact of Urban Climate on Personal Mobility Choices: A Case Study of Seoul’s High-Density Environment
The project aims to explore patterns between the mobility data of 22 participants and climate parameters over a 7-week period in Seoul, South Korea. The project will analyse spatial data of the participant’s locations, using Graph Neural Networks (GNNs) to identify quantifiable mobility patterns. It will then examine potential correlations between the identified patterns and external climate parameters, gathered from weather stations in the city and remote sensing datasets. Secondary objectives include evaluating the effectiveness of GNNs in detecting these patterns and comparing the project’s findings with participants’ self-reported climate preferences and climate comfort models. This latter comparison will help determine whether mobility patterns can offer insight on human climate comfort levels.
Supervisors: Azarakhsh Rafiee and Martín Mosteiro Romero

Noah Petri Alting
Urban Geome-Trees: Automated Modeling of Tree Species and Geometry for CFD in Urban Environments
This thesis aims to support Computational Fluid Dynamics (CFD) simulations of urban wind flow by incorporating detailed representations of vegetation, particularly trees. Currently, trees are either excluded or overly simplified in CFD models, despite their influence on an airflow, such as changing wind speed, the creation of wakes and turbulence patterns. Including detailed tree models can help improve the capabilities of prediction of current CFD techniques. To address this, the thesis is structured around two primary objectives: Automatic Retrieval of Tree Geometries: Developing a method to accurately identify and extract the physical structures of individual trees in different levels of detail using airborne lidar data. This step aims to capture the spatial characteristics of trees to represent them more effectively in CFD models. Automated Identification of Tree Species: Leveraging both lidar and satellite data to automatically classify trees according to their species. By identifying species, this approach is expected to refine the physical characteristics of trees. The simulation parameters assigned to the tree model can then be modelled based on the species information, enhancing the realism of their influence within CFD simulations. By automating these two aspects, this research aims to create an efficient pipeline that enables CFD modellers to easily incorporate realistic tree data, ultimately leading to more accurate urban wind flow simulations.
Supervisors: Hugo Ledoux and lara Garcia-Sanchez

Qiaorui Yang
Building an Intelligent Spatial Knowledge Graph for Energy Management
We aim to create a dynamic, spatially aware knowledge graph that manages interconnected data for energy management, focusing on energy transition planning in the Netherlands. This involves integrating various data types, such as building attributes and renewable energy generation and consumption patterns, into a cohesive, queryable framework.
Supervisors: Prof.dr.ir. Peter van Oosterom, Drs. Wilko Quak and Amin Jalilzadeh

Rafal Marek Tarczynski
Integrating Machine Learning to Enhance Transportation Forecasting in Urban Strategy’s Digital Twin Ecosystem With Graph Neural Networks
This research explores the use of machine learning techniques—particularly Graph Neural Networks (GNNs)—in combination with the Urban Strategy platform to enhance our understanding of the relationship between city morphology and transportation networks. By leveraging the unique spatial and networked properties of GNNs, this study aims to model the impact of urban tissue morphology on transportation flow patterns. The study will involve training machine learning models on existing data from the Rotterdam-The Hague Metropolitan Area and Amsterdam, focusing on how these cities' unique morphological structures influence traffic and freight flow. These models will then be applied to new urban areas to predict freight transportation flows based on their own distinct morphological characteristics. This approach aims to reveal the influence of urban structure on freight dynamics and provide predictive insights for city planners when evaluating transportation flow under varying morphological scenarios. By combining GNNs with MASS-GT modelling capabilities and with Urban Strategy’s multi-modal datasets, this research seeks to build a scalable model that links urban morphology and transportation flows, ultimately enhancing predictive capabilities across diverse urban contexts.
Supervisors: Martijn Meijers and Azarakhsh Rafiee
(company involved: TNO Finn Winkelmann)

Shawn Roy How Wei Tew
Investigating Predictive Modelling for Green Infrastructure Planning and Management
This study investigates the role of green infrastructure in enhancing urban resilience through predictive modeling for land use planning. Green infrastructure, including urban parks, forests, and linear street trees, are essential in enhancing environmental stability and fostering social aspects such as accessibility and supportive policies in urban areas. This research provides insights into how factors such as accessibility, spatial distribution of different green space typologies contributes to urban resilience, specifically the cooling of urban microclimates, managing stormwater, and accessibility to green spaces. Using spatial data and machine learning techniques, this study evaluates different modeling approaches to predict future changes in green space availability under varying environmental scenarios. The findings aim to support urban planners in making data-driven decisions to prioritise and optimise green spaces, thereby contributing to sustainable and resilient urban landscapes. This focused approach to green spaces offers a practical framework for cities seeking to strengthen ecological and community resilience through strategic land use planning.
Supervisors: Daniele Cannatella and Claudiu Forgaci

Victoria Tsalapati
Exploring the impact of tree original point cloud data leverage in urban daylight simulation
Daylighting is a key consideration in urban design contributing to the development of daylight simulations in the past few decades. However, the impact of urban trees is often not considered. However, in cases of 3D model reconstruction that include trees, the simplification of the past attempts to model trees overlook tree canopy porosity. Additionally, throughout the year, the seasonal changes on tree canopies are also often disregarded. This thesis topic revolves around the investigation of the method that integrates of point cloud data into daylighting simulations so that the porosity and the seasonal alterations that characterize the trees in the study area are considered. For this purpose, tree point clouds of diverse density will be used related to the seasonal changes and finally the results will be compared to existing reconstructed 3D tree models. The study area of this thesis is Green Village at TU Delft Campus. As a first approach, through literature research or assumptions it will be defined for which days in the year data should be collect for processing. Then, data of various kinds will be collected, including Illuminance sensor data of CCC Building in Green Village, AHN 4-point cloud data and data from in-situ laser scanning. Next, daylight simulation will be implemented with the resulted point clouds or them after some processing as inputs with either Climate Studio or custom code using Ladybug and Honeybee libraries. Lastly, the results will be evaluated.
Supervisors: Azarakhsh Rafiee and Eleonora Brembilla

Vidushi Bhatt
Geospatial Analytics from Integrate IoT ecosystem in Built spaces
In this thesis I plan to investigate the applications of geospatial analytics within built-space IoT systems to enhance living conditions. The research aims to design a system that examines an indoor space, focusing on parameters like temperature, humidity, and air quality, to gather meaningful data about the indoor environment. Drawing inspiration from available thesis topic, “Netatmo Sensor Integration in IFC model”, this project will employ an open, interoperable framework using the OGC SensorThings API and FROST server for real-time data collection and visualization. This setup allows seamless data flow between various IoT sensors and a PostgreSQL/PostGIS database for geospatial data management. By simulating the system in a test environment, the project will highlight the potential for spatial insights, like energy usage optimization and air quality management. This geospatial analysis not only provides insights into individual living conditions but also sets a foundation for broader applications in building performance and smart urban development.
Supervisors: Azarakhsh Rafiee and Justin Schembri

Walter Hugo Johannes Kahn
Working towards oblique aerial adjustment through the creation of synthetic test cases based on the key adjustment parameters
Adjustment theory is a statistical process to detect and remove stochastic errors from measurements. This is being applied to aerial nadir imagery to help in the reduction of outside surveying for the mapping of the Basis regristratie Grootschalige Topografie (BGT). Unlike nadir images, currently obliques are not adjusted due to difficulties that stem in the trustworthiness of the control points in the image because of a varying pixel scale when compared to top down nadir. However, oblique images do provide extra data opportunities that would otherwise go lost for municipalities that want to cartograph the BGT or maybe even create digital twin. For this it is essential to keep the structure of the image the same during the adjustment process and thus not to morph it or overlay on any surface. Due to the strict precision that the kadaster holds to measurements in the BGT a raw unadjusted oblique image simply does not meet the standards laid out in the Handboek Technische Werkzaamheden (HTW) for the creation of the BGT. This leads us to the goal of the thesis which is to create a series of mathematically defined synthetic test cases that show the impact of key parameters as they relate to adjustment theory. A possible result of this could be that the key points are too inaccurate or that obscuring of objects in different angles leads to a bad fit. The lessons learned from such test cases could then be applied to real data to see whether they hold up in real data scenarios.
Supervisors: Edward Verbree and Martijn Meijers
(company involved: GeoDelta, Annemieke Verbraeck)

Xiaduo Zhao
Reconstructing Complete Roof Geometries from Sparse and Incomplete Point Cloud Data
Roof shape reconstruction into 3D models from airborne LiDAR point clouds remains challenging due to occlusions, artifacts, and sparsity issues in large-scale scanning. While existing methods have made progress in general point cloud completion, they face significant limitations when applied to roof reconstruction, particularly in preserving fine architectural details, maintaining sharp edges, and ensuring accurate scale alignment. These challenges are especially pronounced in complex buildings with multiple roof planes and irregular structures, where current approaches often result in loss of small-scale features, over-smoothing of sharp edges, and normalization inconsistencies. This thesis aims to develop an automated method for reconstructing accurate and detailed 3D roof models from incomplete LiDAR point cloud data. Leveraging recent advances in machine learning, I will explore approaches such as generative models, multi-modal learning, and point cloud densification using Gaussian splatting, selecting the most promising method for further refinement. The expected outcome is a efficient pipeline capable of handling complex and detailed roof geometries ensuring accuracy, sharpness, and scale consistency. The AHN dataset and 3D BAG building models serve as both training and validation data.
Supervisors: Weixiao Gao and Hugo Ledoux

Xiaoluo Gong
Seamless Oblique Image Mosaics for Aerial Visualization
This thesis project, focuses on enhancing oblique aerial imagery by developing techniques for smooth image transitions. Conducted at Geodelta, it aims to create a continuous viewing experience by stitching oblique images seamlessly, useful for both visualization and measurement. Key goals include minimizing visual artifacts in overlapping images captured along a flight path and creating a smooth and continuous transition between consecutive images as well as maintaining the accuracy the stitched images. There are two methods to be applied to achieve those goals. The first method, just find the tie points between two images, interpolate to get the in-betweens, and then morph the rest of the images based on those interpolated points. The other method is using the 3D locations in the point cloud to create tie-points, project the point-cloud points to get the in-between images, and finally do the morph. Both methods will get the interpolating virtual images and require the technique of morphing. This project will be evaluated based on the interpolation errors, the measurement accuracy, and the processing time. By advancing oblique image mosaicking, this project has valuable applications in interactive visualization of landscapes, helping stakeholders gain insights into spatial and structural details within a fluid, navigable environment.
Supervisors: Martijn Meijers and Edward Verbree
(company involved: Annemieke Verbraeck (Geodelta))

Xueheng Li
3D Visualization of Property Valuation information: An LADM Part 4-Compliant Framework for the Netherlands
Property valuation in the Netherlands requires effective dissemination of spatial and temporal data across multiple scales. While the current system manages basic property assessment, it lacks capabilities for multi-level visualization and temporal analysis of valuation data. This research aims to develop an LADM Part 4-compliant prototype system for visualizing and analyzing property valuation information across different spatial levels (apartment, floor, building, street, district) and time periods. The study will utilize cadastral data from a municipality which is willing to share 10 years of valuation history and maybe collaborate with the Waarderingskamer (Netherlands Council for Real Estate Assessment) to create a comprehensive visualization framework. The research will extend the Netherlands LADM Valuation Information Model Country Profile with key objectives: • Designing a multi-level data dissemination structure for property valuations • Developing methods for sketching 3D apartment geometries within buildings • Implementing time-series visualization of WOZ values • Creating animated 3D visualizations across spatial scales • Building an LADM Part 4-compliant DBMSThe The expected outcome is an interactive visualization system that enables analysis of property valuations across spatial scales and time periods. This will improve transparency and support stakeholder decision making through enhanced data communication and temporal analysis capabilities.
Supervisors: Peter van Oosterom and Abdullah Kara

Yan Gao
Labeling vario-scale maps
Effective labeling is important for map readability and usability, yet placing labels dynamically remains a challenge for vario-scale maps. This research will look into optimal label placement and gradual adjustments for vario-scale maps across different scales. One key focus of the research includes defining an optimal label position that adapts to complex and irregular features. Labels will be tested as simple rectangles and as characters aligned along curved features (e.g., for rivers), with repeated labels considered for lengthy features like streets. As the scale changes continuously, overlapping labels may become unavoidable. To maintain readability and spatial alignment with features, this research will also explore dynamic adjustments in label size and placement, as well as gradually removing less important or space problematic labels. A space-scale pyramid (SSP) model will help refine label sizes relative to the screen. Additionally, methods for gradual adjustments will be explored to ensure continuous transitions for labels as scale changes, avoiding jarring movements like popping or sudden shifts. Furthermore, the research may limit to English characters and location names, and may also consider label prioritization strategies to determine which information should (dis)appear or fade out/in(grow/shrink) as scale changes, enhancing map usability. Results may contribute to vario-scale map-building software.
Supervisors: Martijn Meijers and Peter van Oosterom

Zhuoyue Wang
Enhancing Smart Point Clouds with HBIM Attributes for Dynamic Heritage Conservation
This research integrates Historic Building Information Modeling (HBIM) with smart point clouds (SPC) to advance the conservation and management of heritage structures. HBIM, leveraging detailed data about heritage buildings, enhances both structural health monitoring and decision-making in conservation projects. The study aims to enrich SPC with HBIM attributes, creating a dynamic update system that actively manages and monitors heritage sites. Using terrestrial laser scanning and photogrammetry, the project will capture detailed site data, segment point clouds, and apply machine learning for refined tagging, ensuring the HBIM accurately reflects the current state of the site. This integration facilitates ongoing preservation efforts by maintaining a real-time, enriched model of heritage structures.
Supervisors: Peter van Oosterom and Yingwen Yu