Search

In this paper, a model has been created for the automated generation of DTM (Digital Terrain Models) using a hybrid data structure based on LiDAR (Light Detection and Ranging) data. The automated process facilitates the creation and implementation of both simplified and complex DTMs using hybrid data structure. The classification of terrain into various classes is achieved through the application of the slope parameter. This model can be applied to the entire area or a specific area of interest by loading the boundaries and offers the optional capability to implement structural terrain lines in the form of watersheds. The proposed model addresses the challenge of generating DTMs in rapidly changing terrain and areas with complex landscapes, providing solutions to enhance DTM generation performance. Alongside the hybrid DTM structure, the developed model introduces a hierarchical grid with three different spatial resolutions, allowing users to adjust them based on specific requirements. The model is designed to adapt to the terrain, creating a hybrid data structure that combines the benefits of the GRID and TIN (Triangulated Irregular Network) approaches. The development and implementation of the model were carried out in ModelBuilder, within the ArcMap software environment.

This project explores the application of the Apple iPad Pro and Apple iPhone equipped with LiDAR technology using free apps that utilize the LiDAR sensor in a more traditional Geomatics Engineering method. It relies heavily on the IMU (inertial measurement unit), the integrated camera, and the single frequency GNSS (global navigation satellite system) sensor for positioning of the device to control the resultant collected data sets. A terrestrial LiDAR scanner, Leica P20, is used to produce the base surface model for comparison.

Poor road conditions affect millions of drivers, and assessing the condition of paved surfaces is a critical step towards repairing them. This project explores the feasibility of using the Apple iPad Pro LiDAR sensor as a cost-effective tool for assessing the damage and condition of paved surfaces. Our research aims to provide accurate and precise measurements using readily available consumer devices and compare the results to state-of-the-art equipment. This investigation involved visual inspection, identification, and classification of pavement distresses, followed by a comparison of the iPad and iPhone LiDAR data with a survey-grade terrestrial laser scanner. The project revealed several limitations of the iPad Pro-based LiDAR approach. The level of detail captured in the scans was relatively low, with a best-case resolution of 1 cm and an inability to detect smaller cracks and shallow potholes. Longer scans (in terms of both time and distance) led to geometric anomalies in the surface models. Colorized scans provided some visual contrast, aiding in the identification of damage, particularly on moderately damaged concrete surfaces. The potential sources of error were identified, including the performance of the Inertial Measurement Unit (IMU), the limitations of the LiDAR sensor itself, and the opaque nature of onboard data processing within the 3D Scanner App. Suggestions for improvement included the use of gimbal stabilizers to enhance scan quality and the exploration of more intensive PC-based processing for raw data analysis. Hardware advancements by Apple and software enhancements by app developers were also highlighted as potential areas for future improvement. While the project revealed limitations and challenges, the authors acknowledge the possibility of future hardware upgrades, augmented reality advancements, and improvements in sensor accuracy and processing. However, based on this project’s findings, the iPad Pro LiDAR approach currently falls short

Mechanical, electrical, and plumbing (MEP) systems play a crucial role in providing various services and creating comfortable environments for urban residents. In order to enhance the management efficiency of these highly complex MEP systems, as-built building information models (BIMs) are being increasingly adopted worldwide. As-built BIMs accurately represent the actual conditions of facilities, making as-built BIM reconstruction significantly important for construction progress tracking, quality assurance, subsequent facility management, and renewal. To create as-built BIMs for MEP systems, laser scanners are widely utilized to capture high-resolution images and dense 3D measurements of the environment in a fast and highly accurate manner. Despite research efforts to automatically achieve “Scan-to-BIM,” there are still gaps in applying current solutions to real-world scenarios. One of the major challenges are the limited generalization of existing methods to unseen scenarios without proper training or fine-tuning on custom-designed datasets. To address this issue, this study introduces Omni-Scan2BIM, a novel approach powered by large-scale pre-trained vision foundation models. Omni-Scan2BIM enables the recognition of MEP-related components with a single shot by integrating an all-purpose feature extraction model and a class-agnostic segmentation model. Firstly, given only a single image with a reference mask, the visual features are extracted for both the target component and the collected on-site images using the vision foundation model DINOv2. Secondly, through comparing pixel features, similarity maps are generated for the on-site images. The prior points for the class-agnostic segmentation model are sampled from the local maxima of the similarity map. Thirdly, Segment Anything Model (SAM) is leveraged to sequentially segment the target component. Finally, the target component is segmented out in 3D space based on the transformation matrix describing the spati

Digital twin (DT) is capable of managing, simulating, and future performance prediction of existing infrastructures by integrating three-dimensional (3D) geometric model and finite element (FE) model. This study aims to automatically generate geometric DT of existing RC members using point clouds and incorporate damages detected from images into FE models. An edge-based modeling method was proposed to extract sharp features for geometry reconstruction with incomplete point clouds, and an automated procedure was established for registering images to point clouds, mapping cracks detected in images to FE models, and assigning reduced material properties. Four RC beams were tested in a laboratory to validate the proposed method. Existing cracks were found to have a significant effect on structural capacity and failure mode. The results also showed that the developed FE model produced a good agreement between experimental and predicted failure loads while providing a more efficient and time-saving assessment protocol. © 2024

The PointNet architecture is a foundational deep learning model for 3D point clouds, solving classification and segmentation tasks. We hypothesize that the full potential of PointNet has not been reached and is greatly restrained by a single Max pooling layer. First, this thesis introduces new and more complex learnable aggregation functions. Secondly, a novel normalization technique, Context Normalization, is proposed for further feature extraction. Context Normalization is similar to Batch Normalization but independently normalizes each point cloud within a mini-batch and always uses dynamic statistics. The experiments show that replacing Max pooling with Principal Neighborhood Aggregation (PNA) increased classification accuracy from 73.3% to 78.7% on an SO(3) augmented version of the ModelNet40 dataset. Combining PNA with Context Normalization further increased accuracy to 84.6%.

3D-data från både inomhus- och utomhusmiljöer är viktigt för bland annat olika byggsektorsapplikationer, inomhusnavigering och fastighetsförvaltning. Terrester laserskanning (TLS) används för att samla in data för olika ändamål, från terrängkartläggning till skapande av 3D-modeller och byggnadsinformation (BIM). Det är dock viktigt att komma ihåg att ingen laserskanner passar alla situationer, eftersom olika skanningsarbeten kräver olika egenskaper som räckvidd, noggrannhet och detaljnivå. Dessutom kan olika metoder användas beroende på vilket instrument som används och hur skanningsprojektet utformas. Denna studie syftar till att jämföra två laserskanningsmetoder, den ena metoden med laserskannern Trimble X7 och den andra med multistationen Trimble SX12, för att undersöka eventuella skillnader i punktmoln och georeferering. Studien besvarar frågor om osäkerheter i georeferering, avvikelser från noggrant inmätta referenskoordinater, visuella skillnader mellan punktmoln samt för- och nackdelar med metoderna. Två georefereringsmetoder användes i studien, indirekt och direkt georeferering. Studien utfördes i en kontrollerad miljö i en laborationshall (som vanligtvis används för mätningslaborationer) inomhus i Hus 45 Heimdall vid Högskolan i Gävle. För att kunna genomföra studien utfördes förarbete, som bland annat inkluderade nätutjämning i höjd av ett byggnät (som redan finns etablerat i plan) i laborationshallen, markering av kontrollpunkter och utgångspunktermed måltavlor, som även mättes in noggrant med totalstation och helsatsmätning. Skanningsarbetet utfördes med de två instrumenten genom att skanna in väggar i laborationshallen. Efterarbetet omfattades av indirekt georeferering i Trimble RealWorks och en visuell jämförelse av punktmoln i CloudCompare. Vidare gjordes beräkningar för varje metod där RMS-värden,standardavvikelser och medelavvikelser beräknandes. Slutligen gjordes enhypotesprövning i form av ett F-test. Resultaten visade att Trimble SX12 med direkt, 3D data from both indoor and outdoor environments are important for various applications in the construction sector, indoor navigation, and property management. Terrestrial laser scanning (TLS) is used to collect data for various purposes, from terrain mapping to the creation of 3D models and building information modeling (BIM). However, it is important to remember that no single laser scanner is suitable for all situations, as different applications require different characteristics such as range, accuracy, and levelof detail. Additionally, different methods may be used depending on the instrument and the scanning project. This is a study comparing two laser scanning methods: one using the TrimbleX7 laser scanner and the other using the Trimble SX12 multi-station, to investigate potential differences in point clouds and georeferencing. The study addresses questions about uncertainties in georeferencing, deviations from true coordinates, visual differences between point clouds, and the advantages and disadvantages of the methods. Two georeferencing methods were used in the study: indirect and direct georeferencing. The tests were conducted in a controlled environment in a laboratory hall (commonly used for measurements) indoors at Building 45 Heimdall at the University of Gävle. To carry out the study, preliminary work was conducted, which included height network adjustment of the building network (already established in plan) in the laboratory hall, marking control points and starting points with targets, which were also accurately measured with a total station. The scanning work was performed with the two instruments by scanning the walls in the laboratory hall. Post-processing included indirect georeferencingin Trimble RealWorks and visual comparison of point clouds in CloudCompare. Also, RMS, standard deviation, and mean deviation were calculated, and a hypothesis test in the form of an F-test was conducted to obtain the results. The results showed that the Tri

Fotogrammetri är en viktig metod för att skapa 3D-representationer av terräng och strukturer, men utmaningar kvarstår när det gäller noggrannheten på grund av faktorer som bildkvalitet, kamerakalibrering och positionsdata. Användningen av drönare för byggnadsdetaljmätning möjliggör snabb och kostnadseffektiv datainsamling, men noggrannheten kan påverkas av bildkvalitet och skuggning. Avhandlingen syftar till att jämföra noggrannheten och kvaliteten hos punktmoln genererade med två olika tekniker: terrester laserskanning (TLS) och struktur-från-rörelse (SfM) fotogrammetri med drönare. För att testa båda metodernas osäkerhet och noggrannhet vid detaljmätning av bostäder. Genom att utföra mätningar på en villa har data samlats in med både TLS och drönare utrustade med 48 MP kamera, samt georeferering med markstöd (GCP). SfM-punktmoln bearbetades med Agisoft Metashape. Jämförelser gjordes mellan SfM- och TLS-punktmoln avseende täckning, lägesskillnad och lägesosäkerhet. Genom att följa riktlinjer från HMK - Terrester Laserskanning och tillämpa HMK Standardnivå 3 säkerställs hög noggrannhet i mätningarna. Kontroll av lägesosäkerhet av båda punktmolnen resulterade i en lägesosäkerhet som understeg toleranser satta enligt HMK - Terrester laserskanner Standardnivå 3. Kontrollen av lägesosäkerheten visade att kvadratiska medelfelet(RMSE) i plan och höjd var 0.011m respektive 0.007m för TLS-punktmolnet, och 0.02m respektive 0.015m för drönar-SfM-punktmolnet, vilket låg under toleransen enligt HMK- Terrester Detaljmätning 2021. Resultaten tyder på att Structure-from-Motion fotogrammetri med drönare kan generera punktmoln med god detaljrikedom, inte lika noggrann som med terrester laserskanner på sin lägsta inställning. TLS uppvisade mindre osäkerhet enligt kontrollen av lägesosäkerhet, ungefär en halvering av RMSE i både plan och höjd. I studien framgick det att TLS presterar sämre vid svåråtkomliga ytor med skymd sikt och ogynnsamma infallsvinklar, där effekten blir en lägre, Photogrammetry is a crucial method for creating 3D representations of terrain and structures, yet challenges remain regarding accuracy due to factors such as image quality, camera calibration, and positional data. The use of drones for building detail measurements enables rapid and cost-effective data collection, but accuracy can be affected by image quality and shading. This thesis aims to compare the accuracy and quality of point clouds generated using two different techniques: terrestrial laser scanning (TLS) and Structure-from-Motion (SfM) photogrammetry with drones. The objective is to test the uncertainty and accuracy of both methods in residential surveying. Data collection was performed on a villa using both TLS and a drone equipped with a 48 MP camera, along with georeferencing with ground control points (GCP). SfM point clouds were processed with Agisoft Metashape. Comparisons were made between SfM and TLS point clouds in terms of coverage, positional difference, and positional uncertainty. By following guidelines from HMK - Terrester laserskanning 2021 and applying HMK Standard Level 3, high measurement accuracy was ensured. Positional uncertainty checks of both point clouds resulted in positional uncertainty within tolerances set by HMK - Terrestrial Laser Scanning Standard Level 3. The positional uncertainty, with a sample of 41 points showed that the root mean square error (RMSE) in plane and height was 0.011m and 0.007m respectively for the TLS point cloud, and 0.02m and 0.015m for the drone-SfM point cloud, both within the tolerance according to HMK - Terrestrial Detail Measurement 2021. The results suggest that Structure-from-Motion photogrammetry with drones can generate point clouds with good detail, although not as accurate as terrestrial laser scanning at its lowest setting. TLS showed less uncertainty according to the positional uncertainty check, with approximately half the RMSE in both plan and height. The study found that TLS performs worse

Plasma acceleration processes have garnered extensive research interest in recent years due to the versatile applications of high-energy X-rays, including medical diagnostics and treatment, semiconductor manufacturing, and the hardening of material surfaces. However, simulating these processes demands a complex workflow and considerable computational time, rendering in-situ analysis impractical. To mitigate these challenges, simulation-based inference and data-driven methods for one-step inversion can be utilized to recover the matching phase-space representation and elucidate the underlying physics. This work presents the adaptation of a surrogate model specifically for plasma physics processes, such as beam transport in a Free Electron Laser (FEL). A previously proposed model, which uses a mixture of normalizing flows to learn multi-modal distributions, is systematically investigated. Conditionality as a means of surrogate modeling in plasma physics is investigated and trained on simulation results of plasma physics processes. Initial experiments involved using various custom test distributions to derive key insights into the model’s operation. Conditionality was incorporated into the model by utilizing the single-view reconstruction provision appropriately, allowing it to handle a wider range of input conditions. This conditioning has been introduced in two forms, scalar conditioning, and one-hot vector conditioning, and the effectiveness of each method is studied. This redefined architecture was rigorously tested on various test distributions with differing characteristics to validate its performance. The reconstruction quality was then compared to point clouds generated by single normalizing flow-based models with similar network sizes. Additionally, a significant reduction in trainable parameters was achieved, making the model more computationally feasible. Custom training protocols were introduced to further reduce overall training time. The enhanced architec

The objective of this paper is to investigate and propose a method for Indoor Localisation based on Isovists, with the aim of extending the fields of Location-based Services and Geomatics. Various methods and combinations incorporating Isovist concepts, Space Syntax, and visibility graphs are examined and assessed. By investigating these approaches, this study aims to create a comprehensive methodology to achieve localisation using Isovists. The main conclusion drawn from this research is that an Indoor Localisation method based on Isovists is not only feasible but can also effectively support Location-based Services. The analysis and evaluation of all the components have been thoroughly conducted, indicating that when properly integrated, they can provide substantial value for LBS applications. As this is a new method for Indoor Localisation, there is significant scope for future work, particularly in terms of connecting it with existing techniques and integrating them into user applications., Digital Technologies

Environmental collision is a challenging issue in human-robot collaboration. This article proposes a novel fuzzy cluster-based framework for robots to have reactive responses to various environmental collision scenarios. This framework makes four contributions: First, a fuzzy cluster-based environmental collision detection algorithm is developed to efficiently classify the collision area and non-collision (free) area of the environment. Second, based on the collision detection algorithm, a p-norm approximation-based collision avoidance algorithm is proposed to enable robots to avoid environmental collisions with guaranteed stability. Third, by extending the collision avoidance algorithm, an environmental collision adaptation algorithm is proposed to allow robots to adapt to environmental collisions with intelligently regulated contact force. Fourth, a teleoperation controller is designed to strengthen haptic force rendering and enhance the operator’s perception of collisions. Going beyond existing methods, the proposed framework allows teleoperated robots to have real-time responses to collisions in quasi-static environments without suffering from local optima, where the environments can be unstructured, non-convex, and detected with noisy outliers. In addition, this framework is simple in implementation because the proposed collision avoidance and collision adaptation algorithms work as several linear Quadratic Programming (QP) constraints that can be flexibly used by Inverse Kinematics (IK) solvers. Several experiments using 7-Degree of Freedom (DoF) robots are conducted to test and compare the proposed framework with existing methods, demonstrating the effectiveness of our work.

Das Entwurfsthema dieser Masterarbeit ist ein Forschungs- und Präsentationsort für „artifizielles Holz“, eine sich in Entwicklung befindliches neues Bio-Komposit basierend auf Cellulosefasern und Lignin, sowie dessen Anwendung im 3D-Druck.. Das Zentrum soll nicht nur als Forschungseinrichtung, sondern auch als Ausstellungsraum dienen und die Vielseitigkeit des nachhaltigen Materials demonstrieren - das je nach Mischung verschiedene Eigenschaften und Transparenzgrade erhält. Die Konzeption des Zentrums umfasst einen Showroom zur Darstellung von Forschungsergebnissen sowie einen Büroblock, der Seminar- und Werkstatträume beinhaltet. Dies spiegelt die duale Funktion des Zentrums wider: als Ort der Wissensvermittlung und als praktische Werkstatt für Innovationen im Bauwesen. Ein wesentlicher Aspekt des Projekts ist die Integration von nachhaltigen Materialien und Technologien, insbesondere die Verwendung von Holz als nachhaltiger Rohstoff und die Erforschung des Potenzials des 3D-Drucks mit Holz. Die Nachhaltigkeit des Projekts wird durch die ganzheitliche Nutzung von Holz und die Betonung der ökologischen, innovativen und gesellschaftlichen Bedeutung der Holzinitiative hervorgehoben. Die Masterarbeit untersucht, wie traditionelle Holzverarbeitung durch innovative Ansätze ergänzt werden kann, um die Effizienz und Nachhaltigkeit des Bauens zu erhöhen. Im Bereich Design und Konzeptentwicklung wird die Anwendung von Punktwolken als innovative Designmethode vorgestellt. Diese Methode ermöglicht es, Baumaterialien in ihrer Gesamtheit zu betrachten und sie in eine digitale Form zu überführen, die neue Möglichkeiten in der Architektur eröffnet. Der Entwurf des Forschungszentrums soll eine neue Formensprache in der nachhaltigen Architektur einführen und gleichzeitig die interne Struktur und den volumetrischen Aufbau von Baumaterialien berücksichtigen. Die Masterarbeit befasst sich zudem mit den technologischen Innovationen und zukünftigen Perspektiven, insbesondere mit der Roll, Julian Höck, Masterarbeit Universität Innsbruck 2024

Die vorliegende Masterarbeit wurde im Rahmen des Forschungsprojektes „Computational Immediay“ (Subprojekt des SFB Advanced Computational Design) verfasst (Wimmer et al., 2022). Daher wurden in der Arbeit zwei Fragestellungen zusammengeführt: Erstens die aus dem SFB heraus formulierte Forschungsfrage und Aufgabenstellung (Konzeptmodellierung im urbanen Kontext mittels volumetrischen Punktwolkenmodellen) und zweitens einer persönlich von der Autorin gestellten Fragestellung (nach der Erzeugung von Spiritualität durch räumliche Mittel unter Ausschluss von kulturell oder religiös geprägten Symbolen oder Archetypen). SFB / SP3 Computational Immediacy Das Subprojekt SP3 beschäftigt sich mit volumetrischen Punktwolken als Design- und Notationsmedium und der Entwicklung von ML-basierten Methoden, die dem Entwerfer assistieren (Bank et al., 2022). Der Fokus liegt dabei auf der Konzeptfindung im Vorentwurf und der Unterstützung des Designers in der kreativ-offenen Ideenfindung. Volumetrische Punktwolken sind im Inneren „gefüllt“, sie erlauben es also den inneren Aufbau eines Volumens oder die inneren Eigenschaften eines Raums (wie z.B. klimatische Bedingungen, Geräuschpegel, Lichtsituationen, ect.) räumlich zu notieren und diese Daten direkt im Entwurf und in der Modellierung zu verwenden. Im Rahmen dieser Masterarbeit wurden daher folgende Aufgaben gestellt: - Erstellen und Testen einer Methode der Notation „volumetrischer“ also räumlicher Daten eines Bauplatzes und eines architektonischen Programms (Kapitel 4.1 und 4.2) - Erarbeiten und Testen verschiedener Darstellungsmethoden der Punktwolkenmodelle um eine „unmittelbare“ Verständlichkeit zu erreichen (Kapitel 4.3) - Direkte Anwendung der räumlichen Daten in der Modellierung einer Baumasse und Darstellung eines nachvollziehbaren Workflows eingebettet in einen Entwurf (Kapitel 5.0) Diese Aufgabenstellung wurde während einer vom FWF geförderten Anstellung erarbeitet und parallel dazu in einer Masterarbeit mit der Entwic, Katrin Stöhr, B.Sc., Masterarbeit Universität Innsbruck 2024

The tensions between prototype iteration and its associated costs, both in terms of time and resource, are identified as a critical challenge in contemporary prototyping practices. Whilst it is acknowledged that increased prototyping correlates to improved products and greater product innovation, the cost of prototype iteration, particularly in the later stages of the development process is a key factor in determining the extent of prototyping that is undertaken, thus the potential for product success. Addressing this challenge, the research presented in this thesis investigates the development of a methodology for prototype remanufacturing in New Product Development (NPD). Remanufacturing, defined as the restoration of used products to a like-new functional state, emerges as a strategic approach with potential to significantly improve the efficiency of prototype iteration. By remanufacturing, an existing prototype could be ’edited’ to embody only the necessary change between design iterations, promising to reduce not only the time and cost associated with prototyping but also aligning with the shift towards sustainable practices in NPD. This thesis therefore aims to investigate the unique challenges and opportunities that remanufacturing presents when applied in the context of prototyping in NPD. The research adopts a methodological framework encompassing: a comprehensive analysis of design change between high-fidelity prototype iterations, exploring the potential benefits of remanufacturing, and the formalisation of remanufacturing strategies for prototype remanufacturing. Central to this research is the development and optimisation of a computational-remanufacturing tool, designed to enhance the computational efficiency of the prototype remanufacturing process. The research addresses key issues such as the fidelity-efficiency trade-off in prototyping and the need for rapid iteration processes, particularly in the later stages of NPD. Through a series of case stud

In this paper, we have demonstrated the application of a graph convolutional neural network for the purpose of object detection in a LiDAR point cloud. In order to encode the point cloud in the most time-effective manner, we make use of a near-neighbors graph with a defined radius. We create a graph convolutional neural network so that we can find out what kind of object and what class each vertex in a graph represents. We design a box merging and scoring operation to reliably combine detections from numerous vertices into a single score, and we offer an auto-registration strategy as a means of reducing the amount of translation errors that occur inside the system. According to the results of our tests using the KITTI benchmark, we are able to draw the conclusion that the method that was suggested achieves competitive accuracy with the point cloud, even beating fusion-based methods in some instances. According to the results of our research, the graph neural network has the potential to become an effective new tool for the detection of 3D objects.

This article describes the process of building passportization and the subsequent application of an information model for the Faculty of Civil Engineering in Brno. Given the high probability that the building documentation for these structures is outdated, updating the information becomes essential for future maintenance, renovation, and grant applications. The purpose of the article is to provide a description of the procedure for building passportization based on laser scanning technology and showcase the results of the scanning process of Building Z.

Robust detection, localization, and tracking of objects are essential for autonomous driving. Computer vision has largely driven development based on camera sensors in recent years, but 3D localization from images is still challenging. Sensors such as LiDAR or RADAR are used to compute depth; each having its own advantages and drawbacks. The main idea of the project is to be able to mix images from the camera and RADAR detections in order to estimate depths for the objects appearing in the images. Fusion strategies can be considered the solution to give a more detailed description of the environment by utilizing both the 3D localization capabilities of range sensors and the higher spatial resolution of image data. The idea is to fuse 3D detections from the RADAR onto the image plane, this requires a high level of synchronization of the sensors and projections of the RADAR data on the required image., Robust detektering, lokalisering och spårning av objekt är avgörande för autonom körning. Datorseende har till stor del drivit utvecklingen baserad på kamerasensorer de senaste åren, men 3D-lokalisering från bilder är fortfarande utmanande. Sensorer som LiDAR eller RADAR används för att beräkna djup; var och en har sina egna fördelar och nackdelar. Huvudtanken med projektet är att kunna blanda bilder från kameran och RADAR-detektioner för att uppskatta djup för de objekt som förekommer i bilderna. Fusionsstrategier kan anses vara lösningen för att ge en mer detaljerad beskrivning av miljön med både 3D-lokaliseringsförmågan hos avståndssensorer och den högre rumsliga upplösningen av bilddata. Tanken är att smälta samman 3D-detektioner från RADAR till bildplanet, detta kräver en hög nivå av synkronisering av sensorerna och projektioner av RADAR-data på den önskade bilden.

Throughout this project, the aim is to develop a program that, based on the processing of three-dimensional point clouds belonging to objects or shapes, achieves the creation of trajectories through route generation algorithms. These trajectories can be subsequently followed by a manipulator robot. The motivation for this project arises as a contribution to the current development of industrial applications involving robotics, where the use of computer vision techniques is increasingly relevant. This implementation can be beneficial in industrial settings such as welding of metal parts, where trajectory planning along the object's surface is required.

Laserskanning har blivit en vanlig metod för dokumentation, övervakning, underhåll och ut­veckling av olika industrimiljöer. Särskilt för inmätning och visualisering av komplexa rörledningar på industrianläggningars tak är punktmoln från laserskanning ett viktigt verktyg för att på ett enkelt sätt hitta potentiella platser för installation av nya rörledningar. Detta examensarbete genomfördes i samarbete med konsultbolaget Swecos mätningsgrupp i Karlstad och undersökte om det är möj­ligt att ersätta punktmoln från terrester laserskanning med punktmoln skapade med flygbilder tagna med UAS (Unmanned Aerial System) för dokumentering av komplexa rördragningar på industritak. Studien genomfördes på ett mindre område (5x25 m) på reningsverket i stadsdelen Sjöstad (Sjö­stadsverket) i Karlstad. Området innehöll omfattande rörledningar i olika dimensioner och andra detaljer såsom rattar, flänsar och gallerluckor. Detta område ansågs därför vara lämpligt att använda för studiens syfte och tillträdet förutsatte dessutom inte omfattande och dyra säkerhetsutbildningar. Undersökningen genomfördes genom markering av ett sextiotal kontrollpunkter som sedan mättes in med totalstation i ett lokalt referenssystem. Bakåtobjekten användes även som fästpunkter för sfäriska måltavlor under laserskanningen. Dessutom mättes även målade markstödsignaler på bet­ongen in i samband med detta för att möjliggöra en georeferering av flygbilderna. Därefter genom­fördes terrester laserskanning inom undersökningsområdet från nio uppställningar med varierande instrumenthöjder samt två UAS-flygningar med flygfotografering från tio respektive 22 m flyghöjd. Efterbearbetningarna började med att etablera ett lokaltreferenssystem, vilket användes för geo­referering av både laserskanningspunkmolnet samt respektive flygfotograferingspunktmoln. De er­hållna lokala koordinaterna för kontrollpunkterna i respektive punktmoln jämfördes gentemot de totalstationsinmätta koordinaterna för att analysera lägesosäkerhet, This study explored the possibility of using aerial photography from Unmanned Aerial Systems (UAS) as a replacement for terrestrial laser scanning in documenting complex pipeline systems on industrial roofs. The research, conducted in collaboration with Sweco's survey group in Karlstad, focused on visual qualities and positional uncertainty in point clouds generated by terrestrial laser scanning and aerial photography. Control points were marked and surveyed using a total station, then terrestrial laser scanning and UAS-aerial photography was performed to generate point clouds. Analysis revealed that the aerial photography at 22 m altitude was not suitable due to unrecognizable control points. However, the aerial photography at 10 m altitude, after excluding certain types of control points, showed improved positional uncertainty. As the desired applications did not require fine-level detail, UAS aerial photography proved to be a suitable alternative for documenting complex pipeline systems on industrial roofs.

Automatic bin picking is a well-known problem in industrial automation and computer vision, where a robot picks an object from a bin and places it somewhere else. There is continuous ongoing research for many years to improve the contemporary solution. With camera technology advancing rapidly and available fast computation resources, solving this problem with deep learning has become a current interest for several researchers. This thesis intends to leverage the current state-of-the-art deep learning based methods of 3D instance segmentation and point cloud registration and combine them to improve the bin picking solution by improving the performance and make them robust. The problem of bin picking becomes complex when the bin contains identical objects with heavy occlusion. To solve this problem, a 3D instance segmentation is performed with Fast Point Cloud Clustering (FPCC) method to detect and locate the objects in the bin. Further, an extraction strategy is proposed to choose one predicted instance at a time. Inthe next step, a point cloud registration technique is implemented based on PointNetLK method to estimate the pose of the selected object from the bin. The above implementation is trained, tested, and evaluated on synthetically generated datasets. The synthetic dataset also contains several noisy point clouds to imitate a real situation. The real data captured at the company ’SICK IVP’ is also tested with the implemented model. It is observed that the 3D instance segmentation can detect and locate the objects available in the bin. In a noisy environment, the performance degrades as the noise level increase. However, the decrease in the performance is found to be not so significant. Point cloud registration is observed to register best with the full point cloud of the object, when compared to point cloud with missing points.

Usage of asphalt mixture with poor gradation will most likely lead to pavement deficiency. There is a growing need for rapid and non-destructive methods to extract pavement aggregate gradation. In this study, a deep learning-based method that utilizes point clouds data for gradation extraction was proposed. Firstly, a data enhancement algorithm along with three data format conversion methods (aligned point cloud, voxel, and depth image) were proposed to preprocess the original collected point clouds. Subsequently, different neural network models were designed for each data format to extract gradation. Finally, a multi-feature fusion network was developed, which using extraction network as the backbone and additional auxiliary information. In the case study, the MAE loss of multi-feature fusion networks with PointNet, Vox-ResNet34 and GoogLeNet-v4 as the backbone respectively achieved 0.202, 0.142 and 0.046 on the test set, which means an estimation accuracy of more than 95% for the pavement aggregate gradation., Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public., Pavement Engineering

In this study, we perform human identification using accumulated radar point clouds in an outdoor scene. We employ PointNet as classification network and explore the impact of adding radars' non-spatial features as input, namely doppler velocity and radar cross section (RCS). Furthermore, we encode time as an additional time identity dimension to each point within the accumulated point cloud. We examine the effects of normalizing the RCS values, canonicalizing the spatial dimensions of the point cloud, as well as normalizing the doppler velocity with respect to this canonicalization. We examine three different PointNet configurations to understand the impact of the TransformNet blocks (T-Net) within the PointNet architecture on our six-dimensional radar data input. We have created a realistic outdoor dataset for training and evaluation purposes. Our approach of using the unnormalized six-dimensional radar data on the PointNet architecture without the two T-Net blocks achieves the highest performance of 73.4 % on our test set., Mechanical Engineering | Vehicle Engineering | Cognitive Robotics

Subjective quality assessment serves as a method to evaluate the perceptual quality of 3D point clouds. These evaluations can be conducted using lab-based or remote or crowdsourcing tests. The lab-based tests are time-consuming and less cost-effective. As an alternative, remote or crowd tests can be used, offering a time and cost-friendly approach. Remote testing enables larger and more diverse participant pools. However, this raises the question of its applicability due to variability in participants' display devices and environments for the evaluation of the point cloud. In this paper, the focus is on investigating the applicability of remote testing by using the Absolute Category Rating (ACR) test method for assessing the subjective quality of point clouds in different tests. We compare the results of lab and remote tests by replicating lab-based tests. In the first test, we assess the subjective quality of a static point cloud geometry for two different types of geometrical degradations, namely Gaussian noise, and octree-pruning. In the second test, we compare the performance of two different compression methods (G-PCC and V-PCC) to assess the subjective quality of coloured point cloud videos. Based on the results obtained using correlation and Standard deviation of Opinion Scores (SOS) analysis, the remote testing paradigm can be used for evaluating point clouds.

[Resumen] El presente articulo pretende comparar nuestro trabajo anterior en segmentación de estructuras reticulares con redes neuronales frente a un algoritmo ad hoc con el mismo propósito. Actualmente, las redes neuronales o la inteligencia artificial son conceptos muy usados y sinónimos de avances y mejoras, pero en determinados casos es posible emplear técnicas más clásicas, fuera del paradigma de la inteligencia artificial para desarrollar el mismo tipo de tareas con resultados muy similares. Para corroborar esta última mención, en el presente artículo se realiza un análisis comparativo de forma cuantitativa y cualitativa entre un algoritmo ad hoc y el mejor modelo de red neuronal en nuestro último trabajo para segmentar estructuras reticulares. Para la implementación del algoritmo se emplean métodos clásicos como Random Sample Consensus (RANSAC) y crecimiento de regiones. Para realizar la comparación de forma cuantitativa se emplean métricas estandarizadas como precision, recall y f1-score. Estas últimas se calcularán sobre una base de datos propia, compuesta por mil nubes de puntos y generada automáticamente en trabajos anteriores. El algoritmo en cuestión esta diseñado expresamente para tal base de datos., [Abstract] This article aims to compare our previous work on segmentation of reticular structures with neural networks against an ad hoc algorithm for the same purpose. Nowadays neural networks or artificial intelligence are widely used concepts synonymous with advances and improvements, but in certain cases it is possible to use more classical techniques, outside the paradigm of artificial intelligence to achieve the same type of tasks with similar results. To corroborate last mention, in this article we perform a quantitative and qualitative comparative analysis between an ad hoc algorithm and the best neural network model in our latest work for segmenting reticular structures. Conventional methods such as Random Sample Consensus (RANSAC) and region growing are used to implement the algorithm. Standardised metrics such as precision, recall and f1-score are used for quantitative comparison. The latter will be calculated on a proprietary dataset, consisting of a thousand point clouds automatically generated in previous work. The algorithm in question is designed specifically for such a database.

Introduction: 3D semantic segmentation of plant point clouds is an important step towards automatic plant phenotyping and crop modeling. Since traditional hand-designed methods for point-cloud processing face challenges in generalisation, current methods are based on deep neural network that learn to perform the 3D segmentation based on training data. However, these methods require a large annotated training set to perform well. Especially for 3D semantic segmentation, the collection of training data is highly labour intensitive and time consuming. Data augmentation has been shown to improve training on small training sets. However, it is unclear which data-augmentation methods are effective for 3D plant-part segmentation. Methods: In the proposed work, five novel data-augmentation methods (global cropping, brightness adjustment, leaf translation, leaf rotation, and leaf crossover) were proposed and compared to five existing methods (online down sampling, global jittering, global scaling, global rotation, and global translation). The methods were applied to PointNet++ for 3D semantic segmentation of the point clouds of three cultivars of tomato plants (Merlice, Brioso, and Gardener Delight). The point clouds were segmented into soil base, stick, stemwork, and other bio-structures. Results and disccusion: Among the data augmentation methods being proposed in this paper, leaf crossover indicated the most promising result which outperformed the existing ones. Leaf rotation (around Z axis), leaf translation, and cropping also performed well on the 3D tomato plant point clouds, which outperformed most of the existing work apart from global jittering. The proposed 3D data augmentation approaches significantly improve the overfitting caused by the limited training data. The improved plant-part segmentation further enables a more accurate reconstruction of the plant architecture.

This dataset contains RGB and multispectral point clouds of the Sierra Calderona Natural Park, obtained by UAV digital aerial photogrammetry.

Unequal deformation of the soil can cause deformation or damage to buildings, like tilted facades or cracks in walls. This research investigates how deformation of a building can be analyzed using Light Detection And Ranging (LiDAR) data. Cyclomedia captures LiDAR data yearly in the Netherlands making it possible to analyze either one or multiple epochs of data. If deformation is monitored, failure can be predicted, and repairs can be performed in time. A subsiding area is found using the Dutch surface motion map. The data points of a single building are fetched from the LiDAR point cloud to analyze the following types of deformation: a difference in the torsion and tilt angle of the facade between two epochs of data; the tilt angle of the facade for a single epoch of data; and local deformation patterns on the facade. The data is segmented before computing the angles and analyzing the local deformation patterns. During segmentation, points that do not correspond to the façade (like a sunshade or windows) are removed. After segmentation, the facade is modeled by fitting a plane using Principal Component Analysis (PCA). The plane parameters (A, B, C, D) are used to determine the torsion and tilt angles. If the torsion or tilt angles are large (above a degree), it is expected the facade has deformed. Local deformation patterns can be analyzed by using a raster containing the distance of the segmented points with an accuracy of 8 centimeters. Besides the point clouds, Cyclomedia also captures 360° panoramic images (Cycloramas). These Cycloramas can be used to explain patterns that are visible in the rasters. For example, objects in front of the building like a bench or a sunshade. This research uses Random Sample Consensus (RANSAC) to segment the data. From the results, it can be concluded RANSAC is not very predictable because random points are taken as input. So, points corresponding to the facade can be removed instead of points corresponding to wi, Geoscience and Remote Sensing

We propose a sim-to-real framework for dexterous manipulation which can generalize to new objects of the same category in the real world. The key of our framework is to train the manipulation policy with point cloud inputs and dexterous hands. We propose two new techniques to enable joint learning on multiple objects and sim-to-real generalization: (i) using imagined hand point clouds as augmented inputs; and (ii) designing novel contact-based rewards. We empirically evaluate our method using an Allegro Hand to grasp novel objects in both simulation and real world. To the best of our knowledge, this is the first policy learning-based framework that achieves such generalization results with dexterous hands. Our project page is available at https://yzqin.github.io/dexpoint.

Class3Dp is a software tool developed for supervised classification of photogrammetric point clouds. It allows the selection of training samples and classification of point clouds according to different features and machine learning models. Point clouds obtained from photogrammetric algorithms (e.g., Structure from Motion) include spectral information recorded in the image acquisition process using RGB, multispectral or hyperspectral sensors. The software extracts spectral and geometric features from the point clouds to be used for point classification.

Early detection of victims in post-disaster environments is an essential task for rescue teams, and effective victim identification will save the greatest number of lives. One of the most significant challenges to carry out these tasks lies mainly in the harsh conditions typical of post-disaster environments such as low light, victims trapped both in rubble and by some materials that make their location difficult (blankets, wood, glass, Etc.) by simple visual inspection. To face this problem, this Chapter describes a robust method for identifying victims and their location within a heat map of the analyzed environment, using specific equipment (Thermal Camera Optris PI640, Microsoft Kinect, and Robot Dr. Jaguar). For this task development, a novel method developed in ROS is proposed based on the processing of thermal and depth images, in such a way that the first ones are used to detect victims through a Convolutional Neural Network and to generate Thermal Local Point Clouds of the environments in combination with depth images. Iterative Closest Point algorithms were used for a subsequent complete thermal map concatenation of the 3D environment using Point Clouds (Thermal Global Point Clouds). The tests to validate this method were carried out indoors (CAR Arena at Centre for Automation and Robotics) and outdoors (Robotics and Cybernetic Laboratory), recreating environments with post-disaster conditions. The main results show an average of 89% in concatenation efficiency to generate Thermal Global Point Clouds. For different scenarios. On the other hand, the detection of victims was carried out with an efficiency greater than 90% for the different proposed scenarios.

Within precision livestock farming, three-dimensional computer vision can improve growth monitoring in cattle management. To investigate the implementation of three-dimensional computer vision in cattle growth management, this systematic review, adhering to the PRISMA 2020 statement guideline, collected 47 eligible studies from the Web of Science database. Studies were analysed separately based on the incrementally encoded titles, and their outcomes were extracted and recorded in a pre-designed form. The survey of outcomes was conducted by using pivot analysis. The results showed that the body measurements assessment task contributed to other kinds of three-dimensional cattle growth tasks. Using Kinect sensors fixed at nadirs to obtain dorsal features was the most frequently applied approach in three-dimensional data acquisition. For three-dimensional data pre-processing, while empty scene subtraction was the most effective approach to removing background from point clouds, clustering and conditional filters were the most adopted functions to eliminate noise. In the discussion, this review provides actual insights into the knowledge of three-dimensional computer vision in cattle growth management, synthesises common considerations within data acquisition, forms a general procedure of data pre-processing, considers the potential of building an automatic and successive three-dimensional multi-task cattle growth monitoring management system, and discusses factors affecting the performance of models for cattle growth management. This review inspires the practice of future three-dimensional computer vision research in cattle growth management and could be extended to other livestock or wild animals.

3D signals have become widely popular in view of the advantage they provide via 3D representations of data by employing a third spatial or temporal dimension to extend 2D signals. Predominantly, 3D signals contain details inexistent in their 2D counterparts such as the depth of an image, which is inherent to point clouds (PC), or the temporal evolution of an image, which is inherent to time series data such as videos. Despite this advantage, 3D models are still underexploited in machine learning (ML) compared to 2D signals, mainly due to data scarcity. In this thesis, we exploit and determine the efficiency and influence of using both multispectral PCs and time-series data with 3D convolutional neural networks (CNNs). We evaluate the performance and utility of these networks and data in the context of two applications from the areas of digital agriculture and medical imaging. In particular, multispectral PCs are investigated for the problem of fusarium-head-blight (FHB) detection and total number of spikelets estimation, while time-series echocardiography are investigated for the problem of myocardial infarction (MI) detection. In the context of the digital agriculture application, two state-of-the-art datasets were created, namely the UW-MRDC WHEAT-PLANT PC dataset, consisting of 216 multispectral PC of wheat plants, and the UW-MRDC WHEAT-HEAD PC dataset, consisting of 80 multispectral PC of wheat heads. Both dataset samples were acquired using a multispectral 3D scanner. Moreover, a real-time parallel GPU-enabled preprocessing method, that runs 1065 times faster than its CPU counterpart, was proposed to convert multispectral PCs into multispectral 3D images compatible with CNNs. Also, the UW-MRDC WHEAT-PLANT PC dataset was used to develop novel and efficient 3D CNNs for disease detection to automatically identify wheat infected with FHB from multispectral 3D images of wheat plants. In addition, the influence of the multispectral information on the detection perfor

We present a fully automatic approach for reconstructing compact 3D building models from large-scale airborne point clouds. A major challenge of urban reconstruction from airborne LiDAR point clouds lies in that the vertical walls are typically missing. Based on the observation that urban buildings typically consist of planar roofs connected with vertical walls to the ground, we propose an approach to infer the vertical walls directly from the data. With the planar segments of both roofs and walls, we hypothesize the faces of the building surface, and the final model is obtained by using an extended hypothesis-and-selection-based polygonal surface reconstruction framework. Specifically, we introduce a new energy term to encourage roof preferences and two additional hard constraints into the optimization step to ensure correct topology and enhance detail recovery. Experiments on various large-scale airborne LiDAR point clouds have demonstrated that the method is superior to the state-of-the-art methods in terms of reconstruction accuracy and robustness. In addition, we have generated a new dataset with our method consisting of the point clouds and 3D models of 20k real-world buildings. We believe this dataset can stimulate research in urban reconstruction from airborne LiDAR point clouds and the use of 3D city models in urban applications, Urban Data Science

Learning from 3D point-cloud data has rapidly gained momentum, motivated by the success of deep learning on images and the increased availability of 3D~data. In this paper, we aim to construct anisotropic convolution layers that work directly on the surface derived from a point cloud. This is challenging because of the lack of a global coordinate system for tangential directions on surfaces. We introduce DeltaConv, a convolution layer that combines geometric operators from vector calculus to enable the construction of anisotropic filters on point clouds. Because these operators are defined on scalar- and vector-fields, we separate the network into a scalar- and a vector-stream, which are connected by the operators. The vector stream enables the network to explicitly represent, evaluate, and process directional information. Our convolutions are robust and simple to implement and match or improve on state-of-the-art approaches on several benchmarks, while also speeding up training and inference., Comp Graphics & Visualisation

The present dataset contains UAV images during the full blooming period of an apple orchard for three consecutive years, 2018, 2019, and 2020. It is directly linked to a research article entitled “Feasibility assessment of tree-level flower intensity quantification from UAV RGB imagery: A triennial study in an apple orchard”. The data collection site was an apple orchard located at Randwijk, Overbetuwe, The Netherlands (51.938, 5.7068 in WGS84 UTM 31U). This dataset aims to support researchers focussing on remote sensing, machine vision, deep learning, and image classification, and the stakeholders interested in precision horticulture and orchard management. It can be used for flowering intensity estimation and prediction, and spatial and temporal flowering variability mapping by using digital photogrammetry and 3D reconstruction

Fuelled by the increase in popularity of virtual and augmented reality applications, point clouds have emerged as a popular 3D format for acquisition and rendering of digital humans, thanks to their versatility and real-time capabilities. Due to technological constraints and real-time rendering limitations, however, the visual quality of dynamic point cloud contents is seldom evaluated using virtual and augmented reality devices, instead relying on prerecorded videos displayed on conventional 2D screens. In this study, we evaluate how the visual quality of point clouds representing digital humans is affected by compression distortions. In particular, we compare three different viewing conditions based on the degrees of freedom that are granted to the viewer: passive viewing (2DTV), head rotation (3DoF), and rotation and translation (6DoF), to understand how interacting in the virtual space affects the perception of quality. We provide both quantitative and qualitative results of our evaluation involving 78 participants, and we make the data publicly available. To the best of our knowledge, this is the first study evaluating the quality of dynamic point clouds in virtual reality, and comparing it to traditional viewing settings. Results highlight the dependency of visual quality on the content under test, and limitations in the way current data sets are used to evaluate compression solutions. Moreover, influencing factors in quality evaluation in VR, and shortcomings in how point cloud encoding solutions handle visually-lossless compression, are discussed.

Social virtual reality (VR) allows multiple remote users to interact in a shared space, unveiling new possibilities for communication in immersive environments. Mediascape XR presents a social VR experience that teleports 3D representations of remote users, using volumetric video, to a virtual museum. It enables visitors to interact with cultural heritage artifacts while allowing social interactions in real time between them. The application is designed following a human-centered approach, enabling an interactive, educating, and entertaining experience.

A common problem in access control is piggybacking. This is when a person without authorized access sneaks closely behind another with access through a door. This thesis seeks to answer whether using radar is a viable solution when attempting to detect piggybacking. Detection will be made by classifying sequences of point clouds generated by the radar, using neural networks. The thesis compares two different placements of the radar, at the side of- and above a door, with an existing camera based piggybacking detection solution. In addition to comparing the results, the development of the model will be described in detail. This includes exploring different architectures for the neural network(s). Moreover, strengths and weaknesses of radar technology, compared to camera technology will be discussed. The results show that all three solutions perform well, with accuracy above 99\% when one or two people are walking normally in frame. When comparing the solutions on more challenging scenarios such as one person carrying a big box or two people hugging while walking, both radar based solutions outperform the camera based solution. In general, slightly better separation between people can be seen in the point clouds generated by the radar placed above the door. This resulted in slightly better performance compared to the placement at the side of the door in certain scenarios. In a world where privacy and integrity is more valued than ever, radar has a big role to play in modern access control solutions. The results from this thesis show that a radar can perform at the same level, and sometimes better than a camera for detecting piggybacking., Piggybacking is when a potentially malicious person sneaks behind a person with access through a secure door. We created a system which can automatically detect this with almost 100% accuracy using machine learning. It works in both real-time and for recorded data, and can make decisions on both single images and entire sequences. While image recognition is a popular field of research, doing so for 3D point clouds, such as the output of a radar, is not as common. For us it posed unique challenges, on top of being really hard to make sense of with the human eye. There already exists camera based technology for detecting piggybacking. We shall evaluate radar as an alternative solution. It has several advantages compared to camera technology, even when not considering technical advantages such as producing three dimensional images. One can not identify a person using radar, therefore radar does not violate integrity in the same way a camera does. On top of this, producing a radar good enough for piggybacking detection is cheaper than producing a camera for the same purpose. The essence of the thesis was to compare our radar solution to an existing camera solution developed by Axis. Additionally two different placements of the radar were compared, at the side of the door and above the door. The result was that our radar based solution performed slightly better than the camera based solution on easier scenarios, such as one or two people walking normally through a door. For these, both radar positions had accuracies over 99%. When considering more difficult scenarios, such as carrying a large box in front of you, the radars outperformed the camera significantly. When comparing the two placements of the radar, the placement above the door came out slightly ahead. From that point of view, it is easier to see whether one or two people are walking in frame. A real world scenario where our radar based solution could be useful would be somewhere where security is not of the hi

This paper introduces a method to automatically estimate vertical and horizontal clearances of highway viaducts and gantries from Mobile Laser Scanner (MLS) point clouds. It is essential to have accurate data on the vertical and horizontal clearances of overhead infrastructure objects along the highway. Accurate clearance data is used for routing oversized transports, infrastructure reconstruction, maintenance and settling legal claims after incidents. The proposed method takes a point cloud of an infrastructure object as input, and as output provides the user with a concise overview of the horizontal and vertical clearances of the object. A point cloud of a highway overpass or gantry is segmented into the different clusters relevant for determining the clearances. The discrete points in these clusters will then be used to approximate their surfaces with B-splines. Subsequently the minimal clearances can be estimated. These clearances are estimated at certain pre-specified locations according to guidelines from the highway authority. The paper also includes a comparison of the inferred clearances from the point clouds with archived measurements performed by third party contractors. For this case study, a Dutch highway section containing 50 gantries and 20 viaducts is selected. Along this stretch of highway the clearances are estimated. The estimated clearances for each structure are then compared with archived in situ measurements. This will give a quantitative analysis of the quality of the estimated clearances. The estimated vertical clearances have an overestimation of 20-30 mm compared to the validation data. The horizontal clearances show a median underestimation of 20 mm., Optical and Laser Remote Sensing

We investigate how PCServe, a web service for disseminating massive point clouds, performs for read-only access (i.e. a visualization application). PCServe is backed by a database model based on Space Filling Curves. By adding a virtual hierarchy of blocks to the database, we can support different visualization applications for retrieval of point cloud data over the web without having to store the data multiple times. This makes expressive access to point clouds over the web possible. We investigate the amount of processing that is needed to create the database model and how well PCServe handles requests from the visualization application. Some suggestions are provided how the current approach can be improved., GIS Technologie

[EN] The growing availability of three-dimensional (3D) data, such as point clouds, coming from Light Detection and Ranging (LiDAR), Mobile Mapping Systems (MMSs) or Unmanned Aerial Vehicles (UAVs), provides the opportunity to rapidly generate 3D models to support the restoration, conservation, and safeguarding activities of cultural heritage (CH). The so-called scan-to-BIM process can, in fact, benefit from such data, and they can themselves be a source for further analyses or activities on the archaeological and built heritage. There are several ways to exploit this type of data, such as Historic Building Information Modelling (HBIM), mesh creation, rasterisation, classification, and semantic segmentation. The latter, referring to point clouds, is a trending topic not only in the CH domain but also in other fields like autonomous navigation, medicine or retail. Precisely in these sectors, the task of semantic segmentation has been mainly exploited and developed with artificial intelligence techniques. In particular, machine learning (ML) algorithms, and their deep learning (DL) subset, are increasingly applied and have established a solid state-of-the-art in the last half-decade. However, applications of DL techniques on heritage point clouds are still scarce; therefore, we propose to tackle this framework within the built heritage field. Starting from some previous tests with the Dynamic Graph Convolutional Neural Network (DGCNN), in this contribution close attention is paid to: i) the investigation of fine-tuned models, used as a transfer learning technique, ii) the combination of external classifiers, such as Random Forest (RF), with the artificial neural network, and iii) the evaluation of the data augmentation results for the domain-specific ArCH dataset. Finally, after taking into account the main advantages and criticalities, considerations are made on the possibility to profit by this methodology also for non-programming or domain experts., [ES] La creciente disponibilidad de datos tridimensionales (3D), como nubes de puntos, provenientes de la detección de la luz y distancia (LiDAR), sistemas de mapeado móvil (MMS) o vehículos aéreos no tripulados (UAV), brinda la oportunidad de generar rápidamente modelos 3D para apoyar las actividades de restauración, conservación y salvaguardia del patrimonio cultural (CH). El llamado proceso de escaneado-a-BIM puede, de hecho, beneficiarse de dichos datos, y ellos mismos pueden ser una fuente para futuros análisis o actividades sobre el patrimonio arqueológico y el construido. Hay varias formas de explotar este tipo de datos, como el modelado de información de edificios históricos (HBIM), la creación de mallas, la rasterización, la clasificación y la segmentación semántica. Este último, referido a las nubes de puntos, es un tema de máxima actualidad no solo en el dominio del PC sino también en otros campos como la navegación autónoma, la medicina o el comercio minorista. Precisamente en estos sectores, la tarea de la segmentación semántica se ha explotado y desarrollado principalmente con técnicas de inteligencia artificial. En particular, los algoritmos de aprendizaje automático (AA) y su subconjunto de aprendizaje profundo (AP) se aplican cada vez más y han establecido un sólido estado de la técnica en la última media década. Sin embargo, las aplicaciones de las técnicas de AP en las nubes de puntos tradicionales son todavía escasas; por tanto, nos proponemos abordar este marco dentro del ámbito del patrimonio construido. Partiendo de algunas pruebas anteriores con la Red Neural Convolucional de Gráfico Dinámico (DGCNN), en esta contribución se presta atención a: i) la investigación de modelos afinados, utilizados como técnica de aprendizaje por transferencia, ii) la combinación de clasificadores externos, como Random Forest (RF), con la red neuronal artificial, y iii) la evaluación de los resultados de aumentación de datos para el conjunto de datos específico d

Point clouds and sets are ubiquitous, unusual and unstructured data-types which present unique problems to machine learning when used as inputs. Since sets are inherently unaffected by permutations, the input space for these problems naturally forms a non-Euclidean space which is oftentimes not even a manifold. Moreover, similar inputs can have wildly varying cardinalities and so the input space is in general infinite-dimensional. Despite these mathematical difficulties, PointNet (Qi et al.) and Deep Sets (Zaheer et al.) form two foundational contributions for deep learning in this area. In this thesis we study the expressive power of such networks. To that end, we prove theorems about their Lipschitz properties, extend them to well-studied infinite-dimensional spaces, and prove new cardinality-agnostic universality results for point clouds. These results completely characterize the approximable functions and so can be used to compare the representational strength of the underlying model classes. In particular, a normalized version of the DeepSets architecture cannot uniformly approximate the diameter function but can uniformly approximate the center-of-mass function whereas PointNet can uniformly approximate the former but not the latter. Additionally, even when limited to a fixed input cardinality, PointNet cannot uniformly approximate the average value of a continuous function over sets of more than two points. We additionally obtain explicit error lower-bounds for this error of approximation and a present a simple geometric method to produce arbitrarily many examples of this failure-mode. Along the way, we also prove various general purpose universal approximation theorems, one of which is for generalized neural networks whose input space is a set of unknown or nonexistent topology.

[ES] La tecnología de escaneo láser 3D puede obtener rápidamente coordenadas 3D precisas de la superficie delmodelo, pero también tiene algunas deficiencias para la construcción de modelos 3D. En primer lugar, la información de color obtenida no cumple los requisitos de la construcción de un modelo realista; en segundo lugar, los datos de la nube de puntos 3D obtenidos por escaneoamenudoparecenestarhuecos.Lafotogrametríadigitalesunbuenremedioparaestas dos carencias, ya que las imágenes pueden adquirirse a voluntad, y los datos de las imágenes pueden obtenerse para zonas específicas según sea necesario, mientras que la información de textura contenida es más rica. Cómo combinar estos dos métodos para que se complementen con el fin de construir un modelo 3D más detallado es una cuestión que merece la pena investigar. Esta tesis utiliza escáneres láser 3D y fotogrametría digital para obtener datos 3D y datos fotográficos del terreno experimental, y procesa y analiza los datos de la nube de puntos sin procesar mediante el software pertinente y los algoritmos de post-procesamiento de datos relacionados para explorar la teoría y los métodos de aplicación de las técnicas de escáner láser 3D y fotogrametría en la topografía práctica del terreno., [EN] 3D laser scanning technology can quickly obtain accurate 3D coordinates of the model surface, but it also has some shortcomings for the construction of 3D models. Firstly, the colour information obtained does not meet the requirements of realistic model construction; secondly, the 3D point cloud data obtained by scanning often appears to be hollow. Digital photogrammetry is a good remedy for these two shortcomings, as the images can be acquired at will, and the image data can be obtained for specific areas as required, while the texture information contained is richer. How to combine these two methods so that they can complement each other in order to construct a more detailed 3D model is a question worth investigating. This thesis uses 3D laser scanners and digital photogrammetry to obtain 3D data and photographic data of experimentalterrain, and processes and analyses the rawpoint cloud data through relevant software and related data post-processing algorithms to explore the theory and methods of applying 3D laser scanner and photogrammetry techniques in practical terrain surveying.