Liste der Fachpublikationen

Ergebnisse 1 - 32 von 32
Show publication details
Kuban, Katharina; Kuijper, Arjan [1. Review]; Schufrin, Marija [2. Review]

A Gamified Information Visualization for the Exploration of Home Network Traffic Data

2020

Darmstadt, TU, Bachelor Thesis, 2020

Internet users today are exposed to a variety of cyberthreats. Therefore, it is necessary to make even nonexpert users aware of the importance of cybersecurity. In this thesis, an approach to address this problem was developed based on the User Centered Design process. The development focused on visualizing the users home network data to improve security in private usage. Existing approaches are either focusing on visualizing network data for more experienced users or teaching cybersecurity with gamified solutions. Combining both, the visualization of the data was embedded in a game to motivate the user. A user study was conducted to identify the user requirements. It could be shown that the main reasons for not dealing with cybersecurity and network data are the user’s lack of motivation and the difficulty of the topic. While following information visualization and game design principles, a prototype was implemented based on the user requirements. The prototype was evaluated from the user perspective revealing that the game strengthens general awareness for the communication of devices in one’s home network and makes the topic network data more accessible. Additionally, it was found out that the quality of player experience design is a crucial factor to motivate the user in the context of the presented approach. It should therefore get higher attention in future steps.

Show publication details
Neumann, Kai Alexander; Kuijper, Arjan [1. Review]; Domajnko, Matevz [2. Review]; Tausch, Reimar [3. Review]

Adaptive Camera View Clustering for Fast Incremental Image-based 3D Reconstruction

2020

Darmstadt, TU, Bachelor Thesis, 2020

Photogrammetry, more precisely image-based 3D reconstruction, is an established method for digitizing cultural heritage sites and artifacts. This method utilizes images from different perspectives to reconstruct the geometry and texture of an object. What images are necessary for a successful reconstruction depends on the size, shape, and complexity of the object. Therefore, an autonomous scanning system for 3D reconstruction requires some kind of feedback during acquisition. In this thesis, we present an evaluation of different state-of-the-art photogrammetry solutions to identify which of them is most capable of providing feedback that predicts the quality of the final 3D reconstruction during acquisition. For this, we focused on the open-source incremental reconstruction solutions COLMAP, Alicevision Meshroom and MVE. Additionally, we included the commercial solution Agisoft Metashape to evaluate how it compares against the open-source solutions. While we were able to identify some characteristic behaviors, the accuracy and runtime of all four reconstruction solutions vary based on the input dataset. Because of this, and the fact that all four solutions compute very similar results under the same conditions, our tests were not conclusive. Nevertheless, we chose COLMAP as the back-end for further use as it provided good results on the real dataset as well as an extensive command-line interface (CLI). Based on these results, we introduce an iterative image-based reconstruction pipeline that uses a cluster-based acceleration structure to deliver more robust and efficient 3D reconstructions. The photogrammetry solution used for reconstruction is exchangeable. In this pipeline, images that portray common parts of an object are assigned to clusters based on their camera frustums. Each cluster can be reconstructed separately. The pipeline was implemented as a c++ module and tested on the autonomous robotic scanner CultArm3D®. For this system, we embedded the pipeline in a feedback loop with a density-based Next-Best-View (NBV) algorithm to assist during autonomous acquisition.

Show publication details
Chen, Cong; Kuijper, Arjan [1. Review]; Damer, Naser [2. Review]

Advanced analyses of CrazyFaces attacks on face identification systems

2020

Darmstadt, TU, Bachelor Thesis, 2020

After 5 years in prison, the greedy criminal was released. He never gave up the idea of sinagain. But he didn’t want to spend another 5 years in prison. So he began to summarizethe lessons of his last arrest. Five years ago, he was arrested at a bank because surveillancecameras identified him. This was a bit of a surprise to him, because this clever criminal hadrepeatedly escaped the pursuit of surveillance cameras by changing his facial expressions.After investigating, he learned that the face recognition system in that bank is a differentone. Therefore, his previously trained facial expressions failed. So one new idea comesin his mind now, "can i just find one or more facial expressions that can disable moststate-of-the-art face recognition systems?". To known the end of this story, please read therest of this thesis.

Show publication details

Analysis of Schedule and Layout Tuning for Sparse Matrices With Compound Entries on GPUs

2020

Computer Graphics Forum

Large sparse matrices with compound entries, i.e. complex and quaternionic matrices as well as matrices with dense blocks, are a core component of many algorithms in geometry processing, physically based animation and other areas of computer graphics. We generalize several matrix layouts and apply joint schedule and layout autotuning to improve the performance of the sparse matrix-vector product on massively parallel graphics processing units. Compared to schedule tuning without layout tuning, we achieve speedups of up to 5.5×. In comparison to cuSPARSE, we achieve speedups of up to 4.7×.

Show publication details
Grebe, Jonas Henry; Kuijper, Arjan [1. Gutachten]; Terhörst, Philipp [2. Gutachten]

Anomaly-based Face Search

2020

Darmstadt, TU, Bachelor Thesis, 2020

Biometric face identification refers to the use of face images for the automatic identification of individuals. Due to the high performance achieved by current face search algorithms, these algorithms are useful tools, e.g. in criminal investigations. Based on the facial description of a witness, the number of suspects can be significantly reduced. However, while modern face image retrieval approaches either require an accurate verbal description or an example image of the suspect’s face, eyewitness testimonies can seldom provide this level of detail. Moreover, while eyewitness’ recall is one of the most convincing pieces of evidence, it is also one of the most unreliable. Hence, exploiting the more reliable, but vague memories about distinctive facial features directly, such as obvious tattoos, scars or birthmarks, should be considered to filter potential suspects in a first step. This might reduce the risk of wrongful convictions caused by retroactively inferred details in the witness’ recall for subsequent steps. Therefore, this thesis proposes an anomaly-based face search solution that aims at enabling a reduction of the search space solely based on locations of anomalous facial features. We developed an unsupervised image anomaly detection approach based on a cascaded image completion network that allows to roughly localize anomalous regions in face images. (1) This completion model is assumed to fill in deleted regions with probable values conditioned on all the remaining parts of the face image. (2) The reconstruction errors of this model were used as an anomaly signal to create a grid of potential anomaly locations in a given face image. (3) These grids, in the form of a thresholded matrix, were then subsequently used to search for the most relevant images. We evaluated the respective retrieval model on a preprocessed subset of 17.855 images of the VGGFace2 dataset. The three main contributions of this work are (1) a cascaded face image completion approach, (2) an unsupervised inpainting-based anomaly localization approach, and (3) a query-by-anomaly face image retrieval approach. The face inpainting achieved promising results when compared to other recent completion approaches since we didn’t leverage any adversarial component in order to simplify the entire training procedure. These inpaintings enabled to roughly localize anomalies in face images. The proposed retrieval model achieved a 60% hit rate at a penetration rate of about 20% over a gallery of 17.855 images. Despite the limitations of the proposed searching approach, the results revealed the potential benefits of using the more reliable anomaly information to reduce the search space, instead of entirely relying on the elicitation of detailed perpetrator descriptions, either in textual or in visual form.

Show publication details
Bergmann, Tim Alexander; Kuijper, Arjan [1. Gutachten]; Noll, Matthias [2. Gutachten]

AR-Visualisierung von Echtzeitbildgebung für ultraschallgestützte Leberbiopsien

2020

Darmstadt, TU, Master Thesis, 2020

In dieser Arbeit wird ein Augmented Reality-System zur Anzeige von Ultraschallbildern direkt am Patienten vorgestellt. Die Überlagerung wird mit der Hilfe von optisch durchsichtigen Head­mounted Displays durchgeführt. Die lagegerichtete Darstellung der Ultraschallbilder basiert auf einem externen optischen Trackingsystem, dem NDI Polaris Vicra. Um die korrekte Überlagerung zu gewährleisten, wird das Sichtfeld eines Trägers mittels angepasster Single Point Active Alignment Method bestimmt. Die Lage der Ultraschallbilder relativ zu den Tracking-Markierungen der Ultra­schallsonde wird mit einer angepassten Pivot-Kalibrierung ermittelt. Zum objektiven Testen des Systems wurde ein Träger-Dummy verwendet, der das Sehen eines Trägers durch Kameras simuliert. Die Lage von Tracking-Markierungen im Sichtfeld des Träger-Dummies konnte mit einem RMSE von 1,1480 mm bestimmt werden. Bei den Tests der Überlagerung der Ultraschallbilder über den darin repräsentierten Strukturen erreicht das System einen Dice-Koeffizienten von 88,33 %. Zur besseren Skalierung der Berechnungsdauer mit der Anzahl der verwendeten Geräte wurden Matrixoperatoren für die verwendeten Transformationsmatrizen optimiert. Die Berechnungen werden im Schnitt mehr als dreimal so schnell durchgeführt wie die allgemeine Implementierung der Operatoren. Das System versetzt behandelnde Ärzte in die Lage, Ultraschallbilder lagegerichtet über den darin repräsentierten Strukturen zu betrachten. Die Anzeige der Bilder auf einem externen Monitor wird dadurch überflüssig.

Show publication details

Automatic Procedural Model Generation for 3D Object Variation

2020

The Visual Computer

3D objects are used for numerous applications. In many cases not only single objects but also variations of objects are needed. Procedural models can be represented in many different forms, but generally excel in content generation. Therefore this representation is well suited for variation generation of 3D objects. However, the creation of a procedural model can be time-consuming on its own. We propose an automatic generation of a procedural model from a single exemplary 3D object. The procedural model consists of a sequence of parameterizable procedures and represents the object construction process. Changing the parameters of the procedures changes the surface of the 3D object. By linking the surface of the procedural model to the original object surface, we can transfer the changes and enable the possibility of generating variations of the original 3D object. The user can adapt the derived procedural model to easily and intuitively generate variations of the original object. We allow the user to define variation parameters within the procedures to guide a process of generating random variations. We evaluate our approach by computing procedural models for various object types, and we generate variations of all objects using the automatically generated procedural model.

Show publication details
Drozdowski, Pawel; Rathgeb, Christian; Dantcheva, Antitza; Damer, Naser; Busch, Christoph

Demographic Bias in Biometrics: A Survey on an Emerging Challenge

2020

IEEE Transactions on Technology and Society

Systems incorporating biometric technologies have become ubiquitous in personal, commercial, and governmental identity management applications. Both cooperative (e.g. access control) and non-cooperative (e.g. surveillance and forensics) systems have benefited from biometrics. Such systems rely on the uniqueness of certain biological or behavioural characteristics of human beings, which enable for individuals to be reliably recognised using automated algorithms. Recently, however, there has been a wave of public and academic concerns regarding the existence of systemic bias in automated decision systems (including biometrics). Most prominently, face recognition algorithms have often been labelled as “racist” or “biased” by the media, non-governmental organisations, and researchers alike. The main contributions of this article are: (1) an overview of the topic of algorithmic bias in the context of biometrics, (2) a comprehensive survey of the existing literature on biometric bias estimation and mitigation, (3) a discussion of the pertinent technical and social matters, and (4) an outline of the remaining challenges and future work items, both from technological and social points of view.

Show publication details

Demographic Bias in Presentation Attack Detection of Iris Recognition Systems

2020

With the widespread use of biometric systems, the demographic bias problem raises more attention. Although many studies addressed bias issues in biometric verification, there is no works that analyse the bias in presentation attack detection (PAD) decisions. Hence, we investigate and analyze the demographic bias in iris PAD algorithms in this paper. To enable a clear discussion, we adapt the notions of differential performance and differential outcome to the PAD problem. We study the bias in iris PAD using three baselines (hand-crafted, transfer-learning, and training from scratch) using the the NDCLD-2013 database. The experimental results points out that female users will be significantly less protected by the PAD, in comparison to males.

Show publication details
Krispel, Ulrich; Fellner, Dieter W.; Ullrich, Torsten

Distance Measurements of CAD Models in Boundary Representation

2020

Transactions on Computational Science XXXVI
Lecture Notes in Computer Science (LNCS), Transactions on Computational Science
12060, 12060

The need to analyze and visualize distances between objects arises in many use cases. Although the problem to calculate the distance between two polygonal objects may sound simple, real-world scenarios with large models will always be challenging, but optimization techniques – such as space partitioning – can reduce the complexity of the average case significantly. Our contribution to this problem is a publicly available benchmark to compare distance calculation algorithms. To illustrate the usage, we investigated and evaluated a grid-based distance measurement algorithm.

Show publication details
Becker, Hagen; Kuijper, Arjan [1. Gutachten]; Tazari, Mohammad-Reza [2. Gutachten]

Einführung von Serious-Gaming Techniken in die Digitale Physiotherapie der Zukunft

2020

Darmstadt, TU, Master Thesis, 2020

Im Rahmen dieser Arbeit wurde das Spiel PDDanceCity, welches der Bewegungsförderung dient aber auch kognitive Fähigkeiten trainieren soll, in mehreren Punkten erweitert. Das Spiel befand sich zu Beginn der Arbeit noch im Status eines Proof of Conecpt und ist vor allem für ältere Menschen konzeptioniert worden, in dem sich der Spieler auf einer Tanzfläche physisch bewegen muss, um seine Spielfigur durch ein Labyrinth ans Ziel zu steuern. Die Erweiterungen beinhalten zum einen die Möglichkeit das Spiel in der modernen Physiotherapie einzusetzen. Durch das in der Arbeit umgesetzte Profilverwaltungssystem ist es möglich, dass Physiotherapeuten die Entwicklung eines Patienten beobachten, begleiten und gegebenenfalls die Therapie anpassen können. Weitere Punkte waren die Verbesserung und Automatisierung der Steuereinheit des Spieles sowie die Erarbeitung eines Algorithmus für die Generierung der einzelnen Spielkarten, basierend auf den Einstellungen der Profile der Spieler. Zu Beginn musste die Tanzmatte mit der man das Spiel steuert für jeden Spieler neu eingestellt werden und die Generierung einer Spielkarte war zufällig und bezog sich nicht auf Spielerprofile. Durch diese Arbeit entstand ein Algorithmus, welcher die Spielkarten individuell basierend auf den Einstellungen des jeweiligen Spielers generierte. Auch wurde die Kommunikation der Tanzmatte mit dem Spiel verbessert, sodass zukünftig die Kalibrierung der Tanzmatte für jeden einzelnen Spieler entfällt. Außerdem ist es nun durch Gesichtserkennung möglich sich in sein Spielerprofil einzuloggen. Dies soll die Akzeptanz von älteren Menschen verbessern, da sie durch diese Technologie nicht mit Maus und Tastatur agieren müssen und einfacher und schneller das Spiel starten können. Weiterhin wurde eine Studie in einer Einrichtung für ältere Menschen durchgeführt, um Zusammenhänge zwischen der Fitness und der Spielweise eines Probanden zu untersuchen. Der Fitnesszustand jedes Spielers wurde mittels eines unabhängigen Fitnesstestes ermittelt. Nach dem Spielen von PDDanceCity wurden mit verschiedenen Maschinellen Lernen-Algorithmen die Bewegungsdaten der Probanden ausgewertet. Dadurch können künftig Rückschlüsse auf die Fitness der Probanden abhängig von ihrer Spielweise geschlossen werden.

Show publication details
Kraft, Dimitri; Bader, Rainer; Bieber, Gerald

Enhancing Vibroarthrography by using Sensor Fusion

2020

Proceedings of the 9th International Conference on Sensor Network

International Joint Conference on Computer Vision and Computer Graphics Theory and Applications (VISIGRAPP) <15, 2020, Valetta, Malta>

Natural and artificial joints of a human body are emitting vibration and sound during the movement. The sound and vibration pattern of a joint is characteristic and changes due to damage, uneven tread wear, injuries, or other influences. Hence, the vibration and sound analysis enables an estimation of the joint condition. This kind of analysis, vibroarthrography (VAG), allows the analysis of diseases like arthritis or osteoporosis and might determine trauma, inflammation, or misalignment. The classification of the vibration and sound data is very challenging and needs a comprehensive annotated data base. Current existing data bases are very limited and insufficient for deep learning or artificial intelligent approaches. In this paper, we describe a new concept of the design of a vibroarthrography system using a sensor network. We discuss the possible improvements and we give an outlook for the future work and application fields of VAG.

Show publication details

ExerTrack—Towards Smart Surfaces to Track Exercises

2020

Technologies

The concept of the quantified self has gained popularity in recent years with the hype of miniaturized gadgets to monitor vital fitness levels. Smartwatches or smartphone apps and other fitness trackers are overwhelming the market. Most aerobic exercises such as walking, running, or cycling can be accurately recognized using wearable devices. However whole-body exercises such as push-ups, bridges, and sit-ups are performed on the ground and thus cannot be precisely recognized by wearing only one accelerometer. Thus, a floor-based approach is preferred for recognizing whole-body activities. Computer vision techniques on image data also report high recognition accuracy; however, the presence of a camera tends to raise privacy issues in public areas. Therefore, we focus on combining the advantages of ubiquitous proximity-sensing with non-optical sensors to preserve privacy in public areas and maintain low computation cost with a sparse sensor implementation. Our solution is the ExerTrack, an off-the-shelf sports mat equipped with eight sparsely distributed capacitive proximity sensors to recognize eight whole-body fitness exercises with a user-independent recognition accuracy of 93.5% and a user-dependent recognition accuracy of 95.1% based on a test study with 9 participants each performing 2 full sessions. We adopt a template-based approach to count repetitions and reach a user-independent counting accuracy of 93.6 %. The final model can run on a Raspberry Pi 3 in real time. This work includes data-processing of our proposed system and model selection to improve the recognition accuracy and data augmentation technique to regularize the network.

Show publication details
Stockhause, Simon; Ritz, Harald [Referent]; Bormann, Pascal [Korreferent]

Generierung und Ordnung von Events in verteilten Systemen mit asynchroner Kommunikation

2020

Giessen, Technische Hochschule Mittelhessen, Bachelor Thesis, 2020

Der Trend der serviceorientierten Architekturen schafft das Bedürfnis, die Komplexität von verteilten Systemen fassen zu können. Viele bestehende Werkzeuge nutzen Logs und Metriken, um Schlussfolgerungen aus der Anwendung ziehen zu können. Allerdings bieten diese nur eingeschränkt die Möglichkeit, kausal zusammenhänge Events zu erfassen. In dieser Arbeit werden Konzepte zur Darstellung von Events und deren Ordnung in verteilten Systemen präsentiert. Diese werden in praxisnahen Anwendungen eingesetzt. Es wird gezeigt, inwiefern die erarbeiteten Konzepte die spezifizierten Anforderungen und Ziele erfüllen. Um die Eventgenerierung und ihre anschließende Ordnung zu gewährleisten, wird ein Datenmodell beschrieben. Es werden zwei Prototypen zur Kontextpropagierung vorgestellt. Zudem werden Visualisierungsansätze präsentiert, die die erhobenen Daten in ansprechender Form darstellen können. Die implementierte Kontextpropagierung bieten Erfahrungswerte, die für zukünftige Arbeit genutzt werden kann. Die Visualisierungsformen der Frame Galerie und des dreidimensionalen Flamengraphs bieten neue Perspektiven zur Darstellung von Tracingdaten.

Show publication details

GeoRocket: A scalable and cloud-based data store for big geospatial files

2020

SoftwareX

We present GeoRocket, a software for the management of very large geospatial datasets in the cloud. GeoRocket employs a novel way to handle arbitrarily large datasets by splitting them into chunks that are processed individually. The software has a modern reactive architecture and makes use of existing services including Elasticsearch and storage back ends such as MongoDB or Amazon S3. GeoRocket is schema-agnostic and supports a wide range of heterogeneous geospatial file formats. It is also format-preserving and does not alter imported data in any way. The main benefits of GeoRocket are its performance, scalability, and usability, which make it suitable for a number of scientific and commercial use cases dealing with very high data volumes, complex datasets, and high velocity (Big Data). GeoRocket also provides many opportunities for further research in the area of geospatial data management.

Show publication details
Mueller-Roemer, Johannes Sebastian; Fellner, Dieter W. [1. Gutachten]; Stork, André [2. Gutachten]; Müller, Heinrich [3. Gutachten]

GPU Data Structures and Code Generation for Modeling, Simulation, and Visualization

2020

Darmstadt, TU., Diss., 2019

Virtual prototyping, the iterative process of using computer-aided (CAx) modeling, simulation, and visualization tools to optimize prototypes and products before manufacturing the first physical artifact, plays an increasingly important role in the modern product development process. Especially due to the availability of affordable additive manufacturing (AM) methods (3D printing), it is becoming increasingly possible to manufacture customized products or even for customers to print items for themselves. In such cases, the first physical prototype is frequently the final product. In this dissertation, methods to efficiently parallelize modeling, simulation, and visualization operations are examined with the goal of reducing iteration times in the virtual prototyping cycle, while simultaneously improving the availability of the necessary CAx tools. The presented methods focus on parallelization on programmable graphics processing units (GPUs). Modern GPUs are fully programmable massively parallel manycore processors that are characterized by their high energy efficiency and good priceperformance ratio. Additionally, GPUs are already present in many workstations and home computers due to their use in computer-aided design (CAD) and computer games. However, specialized algorithms and data structures are required to make efficient use of the processing power of GPUs. Using the novel GPU-optimized data structures and algorithms as well as the new applications of compiler technology introduced in this dissertation, speedups between approximately one (10×) and more than two orders of magnitude (> 100×) are achieved compared to the state of the art in the three core areas of virtual prototyping. Additionally, memory use and required bandwidths are reduced by up to nearly 86%. As a result, not only can computations on existing models be executed more efficiently but larger models can be created and processed as well. In the area of modeling, efficient discrete mesh processing algorithms are examined with a focus on volumetric meshes. In the field of simulation, the assembly of the large sparse system matrices resulting from the finite element method (FEM) and the simulation of fluid dynamics are accelerated. As sparse matrices form the foundation of the presented approaches to mesh processing and simulation, GPU-optimized sparse matrix data structures and hardware- and domain-specific automatic tuning of these data structures are developed and examined as well. In the area of visualization, visualization latencies in remote visualization of cloud-based simulations are reduced by using an optimizing query compiler. By using hybrid visualization, various user interactions can be performed without network round trip latencies.

Show publication details
Rasheed, Muhammad Irtaza Bin; Kuijper, Arjan [1. Prüfer]; Burkhardt, Dirk [2. Prüfer]

Name Disambiguation

2020

Darmstadt, TU, Master Thesis, 2020

Name ambiguity is a challenge and critical problem in many applications, such as scienti_c literature management, trend analysis etc. The main reason of this is due to di_erent name abbreviations, identical names, name misspellings in publications and bibliographies. An author may have multiple names and multiple authors may have the same name. So when we look for a particular name, many documents containing that person's name may be returned or missed because of the author's di_erent style of writing their name. This can produce name ambiguity which a_ects the performance of document retrieval, web search, database integration, and may result improper classi_cation of authors. Previously, many clustering based algorithm have been proposed, but the problem still remains largely unsolved for both research and industry communities, specially with the fast growth of information available. The aim of this thesis is the implementation of a universal name disambiguation approach that considers almost any existing property to identify authors. After an author of a paper is identi_ed, the normalized name writing form on the paper is used to re_ne the author model and even give an overview about the di_erent writing forms of the author's name. This can be achieved by _rst examine the research on Human-Computer Interaction speci_cally with focus on (Visual) Trend Analysis. Furthermore, a research on di_erent name disambiguation techniques. After that, building a concept and implementing a generalized method to identify author name and a_liation disambiguation while evaluating di_erent properties.

Show publication details

Neue Normen für biometrische Datenaustauschformate

2020

Datenschutz & Datensicherheit

Dieser Artikel gibt einen Überblick über die neuen, erweiterbaren biometrischen Datenaustauschformate in der Normenreihe ISO/IEC 39794. Diese könnten in ein paar Jahren, nach der erforderlichen Vorbereitungszeit, für biometrische Referenzdaten in langlebigen maschinenlesbaren Reisedokumenten eingesetzt werden.

Show publication details
Tamellini, Lorenzo; Chiumenti, Michele; Altenhofen, Christian; Attene, Marco; Barrowclough, Oliver; Livesu, Marco; Marini, Federico; Martinelli, Massimiliano; Skytt, Vibeke

Parametric Shape Optimization for Combined Additive–Subtractive Manufacturing

2020

JOM: The Journal of The Minerals, Metals & Materials Society

In industrial practice, additive manufacturing (AM) processes are often followed by post-processing operations such as heat treatment, subtractive machining, milling, etc., to achieve the desired surface quality and dimensional accuracy. Hence, a given part must be 3D-printed with extra material to enable this finishing phase. This combined additive/subtractive technique can be optimized to reduce manufacturing costs by saving printing time and reducing material and energy usage. In this work, a numerical methodology based on parametric shape optimization is proposed for optimizing the thickness of the extra material, allowing for minimal machining operations while ensuring the finishing requirements. Moreover, the proposed approach is complemented by a novel algorithm for generating inner structures to reduce the part distortion and its weight. The computational effort induced by classical constrained optimization methods is alleviated by replacing both the objective and constraint functions by their sparse grid surrogates. Numerical results showcase the effectiveness of the proposed approach.

Show publication details

PAVED: Pareto Front Visualization for Engineering Design

2020

EuroVis 2020. Eurographics / IEEE VGTC Conference on Visualization 2020

Eurographics / IEEE VGTC Conference on Visualization (EuroVis) <22, 2020, Norrköping, Sweden>

Design problems in engineering typically involve a large solution space and several potentially conflicting criteria. Selecting a compromise solution is often supported by optimization algorithms that compute hundreds of Pareto-optimal solutions, thus informing a decision by the engineer. However, the complexity of evaluating and comparing alternatives increases with the number of criteria that need to be considered at the same time. We present a design study on Pareto front visualization to support engineers in applying their expertise and subjective preferences for selection of the most-preferred solution. We provide a characterization of data and tasks from the parametric design of electric motors. The requirements identified were the basis for our development of PAVED, an interactive parallel coordinates visualization for exploration of multi-criteria alternatives. We reflect on our user-centered design process that included iterative refinement with real data in close collaboration with a domain expert as well as a summative evaluation in the field. The results suggest a high usability of our visualization as part of a real-world engineering design workflow. Our lessons learned can serve as guidance to future visualization developers targeting multi-criteria optimization problems in engineering design or alternative domains

Show publication details
Terhörst, Philipp; Riehl, Kevin; Damer, Naser; Rot, Peter; Bortolato, Blaz; Kirchbuchner, Florian; Struc, Vitomir; Kuijper, Arjan

PE-MIU: A Training-Free Privacy-Enhancing Face Recognition Approach Based on Minimum Information Units

2020

IEEE Access

Research on soft-biometrics showed that privacy-sensitive information can be deduced from biometric data. Utilizing biometric templates only, information about a persons gender, age, ethnicity, sexual orientation, and health state can be deduced. For many applications, these templates are expected to be used for recognition purposes only. Thus, extracting this information raises major privacy issues. Previous work proposed two kinds of learning-based solutions for this problem. The first ones provide strong privacy-enhancements, but limited to pre-defined attributes. The second ones achieve more comprehensive but weaker privacy-improvements. In this work, we propose a Privacy-Enhancing face recognition approach based on Minimum Information Units (PE-MIU). PE-MIU, as we demonstrate in this work, is a privacy-enhancement approach for face recognition templates that achieves strong privacy-improvements and is not limited to pre-defined attributes. We exploit the structural differences between face recognition and facial attribute estimation by creating templates in a mixed representation of minimal information units. These representations contain pattern of privacy-sensitive attributes in a highly randomized form. Therefore, the estimation of these attributes becomes hard for function creep attacks. During verification, these units of a probe template are assigned to the units of a reference template by solving an optimal best-matching problem. This allows our approach to maintain a high recognition ability. The experiments are conducted on three publicly available datasets and with five state-of-the-art approaches. Moreover, we conduct the experiments simulating an attacker that knows and adapts to the systems privacy mechanism. The experiments demonstrate that PE-MIU is able to suppress privacy-sensitive information to a significantly higher degree than previous work in all investigated scenarios. At the same time, our solution is able to achieve a verification performance close to that of the unmodified recognition system. Unlike previous works, our approach offers a strong and comprehensive privacy-enhancement without the need of training.

Show publication details
Nottebaum, Moritz; Kuijper, Arjan [1. Review]; Rus, Silvia [2. Review]

Person Re-identification in a Car Seat

2020

Darmstadt, TU, Bachelor Thesis, 2020

In this thesis, I enhanced a car seat with 16 capacity sensors, which collect data from the person sitting on it, which is then used to train a machine learning algorithm to re-identify the person from a group of other already trained persons. In practice, the car seat recognizes the person when he/she sits on the car seat and greets the person with their own name, enabling various customisations in the car unique to the user, like seat configurations, to be applied. Many researchers have done similar things with car seats or seats in general, though focusing on other topics like posture classification. Other interesting use cases of capacitive sensor enhanced seats involved measuring the emotions or focusing on general activity recognition. One major challenge in capacitive sensor research is the inconstancy of the received data, as they are not only affected by objects or persons near to it, but also by changing effects like humidity and temperature. My goal was to make the re-identification robust and use a learning algorithm which can quickly learn the patterns of new persons and is able to achieve satisfiable results even after getting only few training instances to learn from. Another important property was to have a learning algorithm which can operate independent and fast to be even applicable in cars. Both points were achieved by using a shallow convolutional neural network which learns an embedding and is trained with triplet loss, resulting in a computationally cheap inference. In Evaluation, results showed that neural networks are definitely not always the best choice, even though the computation time difference is insignificant. Without enough training data, they often lack in generalisation over the training data. Therefore an ensemble-learning approach with majority voting proved to be the best choice for this setup.

Show publication details
Grimm, Niklas; Damer, Naser [1. Gutachten]; Kuijper, Arjan [2. Gutachten]

Poseninvariante Handerkennung mit generativer Bildkorrektur

2020

Darmstadt, TU, Bachelor Thesis, 2020

Auf Händen basierende biometrische Verfahren sind in weiten Bevölkerungsschichten akzeptiert und können kontaktlos angewendet werden. Bei diesen kontaktlosen Authentifizierungsverfahren sind variierende Handposen eines der größten Probleme. Diese Arbeit erforscht, ob es möglich ist, aus Bildern mit variierenden Handposen solche zu synthetisieren, die einer normalisierten Handpose entsprechen. Auf der Grundlage dieser normalisierten Handpose wäre dann ein besserer Vergleich mit dem Referenzbild möglich. Das Fehlen eines großen Datensatzes und die variierende Skalierung bei kontaktlos akquirierten Handbildern bringt viele Herausforderungen. Von diesen Herausforderungen motiviert beschreibt diese Arbeit mehrere Experimente mit einer begrenzten Menge an Trainingsdaten und variierenden Skalierungen der Eingabebilder, echt wirkende Bilder mit normalisierten Handposen, zu synthetisieren. Die synthetisierten Bilder werden in Verifikationsverfahren mit den Orginalbildern verglichen. Am Ende zeigten die Experimente, dass es nicht möglich ist, mit diesem Versuchsaufbau echt wirkende Handbilder mit normalisierten Posen zu synthetisieren.

Show publication details
González, Camila; Kuijper, Arjan [1. Gutachten]; Mukhopadhyay, Anirban [2. Gutachten]

Preventing Catastrophic Forgetting in Deep Learning Classifiers

2020

Darmstadt, TU, Master Thesis, 2020

Deep neural networks suffer from the problem of catastrophic forgetting. When a model is trained sequentially with batches of data coming from different domains, it adapts too strongly to properties present on the last batch. This causes a catastrophic fall in performance for data similar to that in the initial batches of training. Regularization-based methods are a popular way to reduce the degree of forgetting, as they have an array of desirable properties. However, they perform poorly when no information about the data origin is present at inference time. We propose a way to improve the performance of such methods which comprises introducing insularoty noise in unimportant parameters so that the model grows robust against them changing. Additionally, we present a way to bypass the need for sourcing information. We propose using an oracle to decide which of the previously seen domains a new instance belongs to. The oracle’s prediction is then used to select the model state. In this work, we introduce three such oracles. Two of these select the model which is most confident for the instance. The first, the cross-entropy oracle, chooses the model with least cross-entropy between the prediction and the one-hot form of the prediction. The second, the MC dropout oracle, chooses the model with lowest standard deviation between predictions resulting from performing an array of forward passes while applying dropout. Finally, the domain identification oracle extracts information about the data distribution for each task using the training data. At inference time, it assesses which task the instance is likeliest to belong to, and applies the corresponding model. For all of our three different datasets, at least one oracle performs better than all regularization-based methods. Furthermore, we show that the oracles can be combined with a sparsification-based approach that significantly reduces the memory requirements.

Show publication details
Cao, Min; Chen, Chen; Dou, Hao; Hu, Xiyuan; Peng, Silong; Kuijper, Arjan

Progressive Bilateral-Context Driven Model for Post-Processing Person Re-Identification

2020

IEEE Transactions on Multimedia

Most existing person re-identification methods compute pairwise similarity by extracting robust visual features and learning the discriminative metric. Owing to visual ambiguities, these content-based methods that determine the pairwise relationship only based on the similarity between them, inevitably produce a suboptimal ranking list. Instead, the pairwise similarity can be estimated more accurately along the geodesic path of the underlying data manifold by exploring the rich contextual information of the sample. In this paper, we propose a lightweight post-processing person re-identification method in which the pairwise measure is determined by the relationship between the sample and the counterpart's context in an unsupervised way. We translate the point-to-point comparison into the bilateral point-to-set comparison. The sample's context is composed of its neighbor samples with two different definition ways: the first order context and the second order context, which are used to compute the pairwise similarity in sequence, resulting in a progressive post-processing model. The experiments on four large-scale person re-identification benchmark datasets indicate that (1) the proposed method can consistently achieve higher accuracies by serving as a post-processing procedure after the content-based person re-identification methods, showing its state-of-the-art results, (2) the proposed lightweight method only needs about 6 milliseconds for optimizing the ranking results of one sample, showing its high-efficiency. Code is available at: https://github.com/123ci/PBCmodel.

Show publication details
Fina, Kenten; Kuijper, Arjan [1. Review]; Urban, Philipp [2. Review]; Dennstädt, Marco [3. Review]

Real-time rendering of CSG-operations on high resolution data for preview of 3D-prints

2020

Darmstadt, TU, Master Thesis, 2020

In this thesis various optimizations for the ray-marching algorithm are introduced to efficiently render CSG-operations on high resolution meshes. By using a 2- pass render method and CSG-node memory method, speed-ups of factor 2 to 3 can be achieved in contrast to standard ray marching. We implement a oct-tree based data structure to compress the high resolution SDF (signed distance function) as well as color data. For raw data at a resolution 1024^3, our compressed data requires on average 1.69% of the raw data. Lastly we compare our performance against the openCSG implementations of the well-known Goldfeather and SCS algorithm.

Show publication details

Sensing Technology for Human Activity Recognition: a Comprehensive Survey

2020

IEEE Access

Sensors are devices that quantify the physical aspects of the world around us. This ability is important to gain knowledge about human activities. Human Activity recognition plays an import role in people’s everyday life. In order to solve many human-centered problems, such as health-care, and individual assistance, the need to infer various simple to complex human activities is prominent. Therefore, having a well defined categorization of sensing technology is essential for the systematic design of human activity recognition systems. By extending the sensor categorization proposed by White, we survey the most prominent research works that utilize different sensing technologies for human activity recognition tasks. To the best of our knowledge, there is no thorough sensor-driven survey that considers all sensor categories in the domain of human activity recognition with respect to the sampled physical properties, including a detailed comparison across sensor categories. Thus, our contribution is to close this gap by providing an insight into the state-of-the-art developments. We identify the limitations with respect to the hardware and software characteristics of each sensor category and draw comparisons based on benchmark features retrieved from the research works introduced in this survey. Finally, we conclude with general remarks and provide future research directions for human activity recognition within the presented sensor categorization.

Show publication details
Pöllabauer, Thomas Jürgen; Rojtberg, Pavel [1. Prüfer]; Kuijper, Arjan [2. Prüfer]

STYLE: Style Transfer for Synthetic Training of a YoLo6D Pose Estimator

2020

Darmstadt, TU, Master Thesis, 2020

Supervised training of deep neural networks requires a large amount of training data. Since labeling is time-consuming and error prone and many applications lack data sets of adequate size, research soon became interested in generating this data synthetically, e.g. by rendering images, which makes the annotation free and allows utilizing other sources of available data, for example, CAD models. However, unless much effort is invested, synthetically generated data usually does not exhibit the exact same properties as real-word data. In context of images, there is a difference in the distribution of image features between synthetic and real imagery, a domain gap. This domain gap reduces the transfer-ability of synthetically trained models, hurting their real world inference performance. Current state-of-the-art approaches trying to mitigate this problem concentrate on domain randomization: Overwhelming the model’s feature extractor with enough variation to force it to learn more meaningful features, effectively rendering real-world images nothing more but one additional variation. The main problem with most domain randomization approaches is that it requires the practitioner to decide on the amount of randomization required, a fact research calls "blind" randomization. Domain adaptation in contrast directly tackles the domain gap without the assistance of the practitioner, which makes this approach seem superior. This work deals with training of a DNN-based object pose estimator in three scenarios: First, a small amount of real-world images of the objects of interest is available, second, no images are available, but object specific texture is given, and third, no images and no textures are available. Instead of copying successful randomization techniques, these three problems are tackled mainly with domain adaptation techniques. The main proposition is the adaptation of general-purpose, widely-available, pixel-level style transfer to directly tackle the differences in features found in images from different domains. To that end several approaches are introduced and tested, corresponding to the three different scenarios. It is demonstrated that in scenario one and two, conventional conditional GANs can drastically reduce the domain gap, thereby improving performance by a large margin when compared to non-photo-realistic renderings. More importantly: ready-to-use style transfer solutions improve performance significantly when compared to a model trained with the same degree of randomization, even when there is no real-world data of the target objects available (scenario three), thereby reducing the reliance on domain randomization.

Show publication details

Time-unfolding Object Existence Detection in Low-quality Underwater Videos using Convolutional Neural Networks

2020

Proceedings of the 15th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications

International Joint Conference on Computer Vision and Computer Graphics Theory and Applications (VISIGRAPP) <15, 2020, Valetta, Malta>

Monitoring the environment for early recognition of changes is necessary for assessing the success of renaturation measures on a facts basis. It is also used in fisheries and livestock production for monitoring and for quality assurance. The goal of the presented system is to count sea trouts annually over the course of several months. Sea trouts are detected with underwater camera systems triggered by motion sensors. Such a scenario generates many videos that have to be evaluated manually. This article describes the techniques used to automate the image evaluation process. An effective method has been developed to classify videos and determine the times of occurrence of sea trouts, while significantly reducing the annotation effort. A convolutional neural network has been trained via supervised learning. The underlying images are frame compositions automatically extracted from videos on which sea trouts are to be detected. The accuracy of the resulting detection system reaches values of up to 97.7 %.

Show publication details

Towards 3D Digitization in the GLAM (Galleries, Libraries, Archives, and Museums) Sector – Lessons Learned and Future Outlook

2020

The IPSI BgD Transactions on Internet Research

The European Cultural Heritage Strategy for the 21st century, within the Digital Agenda, one of the flagship initiatives of the Europe 2020 Strategy, has led to an increased demand for fast, efficient and faithful 3D digitization technologies for cultural heritage artefacts. 3D digitization has proven to be a promising approach to enable precise reconstructions of objects. Yet, unlike the digital acquisition of cultural goods in 2D which is widely used and automated today, 3D digitization often still requires significant manual intervention, time and money. To enable heritage institutions to make use of large scale, economic, and automated 3D digitization technologies, the Competence Center for Cultural Heritage Digitization at the Fraunhofer Institute for Computer Graphics Research IGD has developed CultLab3D, the world’s first fully automatic 3D mass digitization technology for collections of three-dimensional objects. 3D scanning robots such as the CultArm3D-P are specifically designed to automate the entire 3D digitization process thus allowing to capture and archive objects on a large-scale and produce highly accurate photo-realistic representations. The unique setup allows to shorten the time needed for digitization from several hours to several minutes per artefact.

Show publication details

Transforming Seismocardiograms Into Electrocardiograms by Applying Convolutional Autoencoders

2020

2020 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings

45th International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2020) <45, 2020, Barcelona, Spain>

Electrocardiograms constitute the key diagnostic tool for cardiologists. While their diagnostic value is yet unparalleled, electrode placement is prone to errors, and sticky electrodes pose a risk for skin irritations and may detach in long-term measurements. Heart.AI presents a fundamentally new approach, transforming motion-based seismocardiograms into electrocardiograms interpretable by cardiologists. Measurements are conducted simply by placing a sensor on the user’s chest. To generate the transformation model, we trained a convolutional autoencoder with the publicly available CEBS dataset. The transformed ECG strongly correlates with the ground truth (r=.94, p<.01), and important features (number of R-peaks, QRS-complex durations) are modeled realistically (Bland-Altman analyses, p>0.12). On a 5- point Likert scale, 15 cardiologists rated the morphological and rhythmological validity as high (4.63/5 and 4.8/5, respectively). Our electrodeless approach solves crucial problems of ECG measurements while being scalable, accessible and inexpensive. It contributes to telemedicine, especially in low-income and rural regions worldwide.

Show publication details
Metzler, Simon Konstantin; Kuijper, Arjan [1. Review]; Yeste Magdaleno, Javier [2. Review]

Visually-aware Recommendation System for Interior Design

2020

Darmstadt, TU, Bachelor Thesis, 2020

Suitable recommendations are critical for a successful e-commerce experience, especially for product categories such as furniture. A well thought-out choice of furniture is decisive for the visual appearance and the comfort of a room. Interior design can take much time and not everyone is capable to do it. Some furniture stores offer recommendation systems on their website, which are usually based on collaborative filters that are very restrictive, can be inaccurate and require many data at first. This work aims to develop a method to provide set recommendations that adhere to a cohesive visual style. The method can automatically advise the user on what set of furniture to choose for a room around one seed piece. The proposed system uses a database where learned attributes of the dataset are previously stored. Once the user select a seed, the system extracts the attributes from the image to execute a query in the database. Finally, a visual search performed in the filtered subset will return the best candidates. This way has the advantage to receive the results faster and to reduce the searching space thereby improving efficiency. The system is presented that is both powerful and efficient enough to give useful user-specific recommendations in real-time.