Kaiserslautern - Fachbereich Informatik
Refine
Year of publication
Document Type
- Doctoral Thesis (218) (remove)
Language
- English (218) (remove)
Has Fulltext
- yes (218)
Keywords
- Visualisierung (16)
- Visualization (8)
- Deep Learning (7)
- Computergraphik (5)
- Evaluation (4)
- Artificial Intelligence (3)
- Geoinformationssystem (3)
- Machine Learning (3)
- Mustererkennung (3)
- Optische Zeichenerkennung (3)
Faculty / Organisational entity
Distributed systems are omnipresent nowadays and networking them is fundamental for the continuous dissemination and thus availability of data. Provision of data in real-time is one of the most important non-functional aspects that safety-critical networks must guarantee. Formal verification of data communication against worst-case deadline requirements is key to certification of emerging x-by-wire systems. Verification allows aircraft to take off, cars to steer by wire, and safety-critical industrial facilities to operate. Therefore, different methodologies for worst-case modeling and analysis of real-time systems have been established. Among them is deterministic Network Calculus (NC), a versatile technique that is applicable across multiple domains such as packet switching, task scheduling, system on chip, software-defined networking, data center networking and network virtualization. NC is a methodology to derive deterministic bounds on two crucial performance metrics of communication systems:
(a) the end-to-end delay data flows experience and
(b) the buffer space required by a server to queue all incoming data.
NC has already seen application in the industry, for instance, basic results have been used to certify the backbone network of the Airbus A380 aircraft.
The NC methodology for worst-case performance analysis of distributed real-time systems consists of two branches. Both share the NC network model but diverge regarding their respective derivation of performance bounds, i.e., their analysis principle. NC was created as a deterministic system theory for queueing analysis and its operations were later cast in a (min,+)-algebraic framework. This branch is known as algebraic Network Calculus (algNC). While algNC can efficiently compute bounds on delay and backlog, the algebraic manipulations do not allow NC to attain the most accurate bounds achievable for the given network model. These tight performance bounds can only be attained with the other, newly established branch of NC, the optimization-based analysis (optNC). However, the only optNC analysis that can currently derive tight bounds was proven to be computationally infeasible even for the analysis of moderately sized networks other than simple sequences of servers.
This thesis makes various contributions in the area of algNC: accuracy within the existing framework is improved, distributivity of the sensor network calculus analysis is established, and most significantly the algNC is extended with optimization principles. They allow algNC to derive performance bounds that are competitive with optNC. Moreover, the computational efficiency of the new NC approach is improved such that this thesis presents the first NC analysis that is both accurate and computationally feasible at the same time. It allows NC to scale to larger, more complex systems that require formal verification of their real-time capabilities.
Crowd condition monitoring concerns the crowd safety and concerns business performance metrics. The research problem to be solved is a crowd condition estimation approach to enable and support the supervision of mass events by first-responders and marketing experts, but is also targeted towards supporting social scientists, journalists, historians, public relations experts, community leaders, and political researchers. Real-time insights of the crowd condition is desired for quick reactions and historic crowd conditions measurements are desired for profound post-event crowd condition analysis.
This thesis aims to provide a systematic understanding of different approaches for crowd condition estimation by relying on 2.4 GHz signals and its variation in crowds of people, proposes and categorizes possible sensing approaches, applies supervised machine learning algorithms, and demonstrates experimental evaluation results. I categorize four sensing approaches. Firstly, stationary sensors which are sensing crowd centric signals sources. Secondly, stationary sensors which are sensing other stationary signals sources (either opportunistic or special purpose signal sources). Thirdly, a few volunteers within the crowd equipped with sensors which are sensing other surrounding crowd centric device signals (either individually, in a single group or collaboratively) within a small region. Fourthly, a small subset of participants within the crowd equipped with sensors and roaming throughout a whole city to sense wireless crowd centric signals.
I present and evaluate an approach with meshed stationary sensors which were sensing crowd centric devices. This was demonstrated and empirically evaluated within an industrial project during three of the world-wide largest automotive exhibitions. With over 30 meshed stationary sensors in an optimized setup across 6400m2 I achieved a mean absolute error of the crowd density of just 0.0115
people per square meter which equals to an average of below 6% mean relative error from the ground truth. I validate the contextual crowd condition anomaly detection method during the visit of chancellor Mrs. Merkel and during a large press conference during the exhibition. I present the approach of opportunistically sensing stationary based wireless signal variations and validate this during the Hannover CeBIT exhibition with 80 opportunistic sources with a crowd condition estimation relative error of below 12% relying only on surrounding signals in influenced by humans. Pursuing this approach I present an approach with dedicated signal sources and sensors to estimate the condition of shared office environments. I demonstrate methods being viable to even detect low density static crowds, such as people sitting at their desks, and evaluate this on an eight person office scenario. I present the approach of mobile crowd density estimation by a group of sensors detecting other crowd centric devices in the proximity with a classification accuracy of the crowd density of 66 % (improvement of over 22% over a individual sensor) during the crowded Oktoberfest event. I propose a collaborative mobile sensing approach which makes the system more robust against variations that may result from the background of the people rather than the crowd condition with differential features taking information about the link structure between actively scanning devices, the ratio between values observed by different devices, ratio of discovered crowd devices over time, team-wise diversity of discovered devices, number of semi- continuous device visibility periods, and device visibility durations into account. I validate the approach on multiple experiments including the Kaiserslautern European soccer championship public viewing event and evaluated the collaborative mobile sensing approach with a crowd condition estimation accuracy of 77 % while outperforming previous methods by 21%. I present the feasibility of deploying the wireless crowd condition sensing approach to a citywide scale during an event in Zurich with 971 actively sensing participants and outperformed the reference method by 24% in average.
Wearable activity recognition aims to identify and assess human activities with the help
of computer systems by evaluating signals of sensors which can be attached to the human
body. This provides us with valuable information in several areas: in health care, e.g. fluid
and food intake monitoring; in sports, e.g. training support and monitoring; in entertainment,
e.g. human-computer interface using body movements; in industrial scenarios, e.g.
computer support for detected work tasks. Several challenges exist for wearable activity
recognition: a large number of nonrelevant activities (null class), the evaluation of large
numbers of sensor signals (curse of dimensionality), ambiguity of sensor signals compared
to the activities and finally the high variability of human activity in general.
This thesis develops a new activity recognition strategy, called invariants classification,
which addresses these challenges, especially the variability in human activities. The
core idea is that often even highly variable actions include short, more or less invariant
sub-actions which are due to hard physical constraints. If someone opens a door, the
movement of the hand to the door handle is not fixed. However the door handle has to
be pushed to open the door. The invariants classification algorithm is structured in four
phases: segmentation, invariant identification, classification, and spotting. The segmentation
divides the continuous sensor data stream into meaningful parts, which are related
to sub-activities. Our segmentation strategy uses the zero crossings of the central difference
quotient of the sensor signals, as segment borders. The invariant identification finds
the invariant sub-activities by means of clustering and a selection strategy dependent on
certain features. The classification identifies the segments of a specific activity class, using
models generated from the invariant sub-activities. The models include the invariant
sub-activity signal and features calculated on sensor signals related to the sub-activity. In
the spotting, the classified segments are used to find the entire activity class instances in
the continuous sensor data stream. For this purpose, we use the position of the invariant
sub-activity in the related activity class instance for the estimation of the borders of the
activity instances.
In this thesis, we show that our new activity recognition strategy, built on invariant
sub-activities, is beneficial. We tested it on three human activity datasets with wearable
inertial measurement units (IMU). Compared to previous publications on the same
datasets we got improvement in the activity recognition in several classes, some with a
large margin. Our segmentation achieves a sensible method to separate the sensor data in
relation to the underlying activities. Relying on sub-activities makes us independent from
imprecise labels on the training data. After the identification of invariant sub-activities,
we calculate a value called cluster precision for each sensor signal and each class activity.
This tells us which classes can be easily classified and which sensor channels support
the classification best. Finally, in the training for each activity class, our algorithm selects
suitable signal channels with invariant sub-activities on different points in time and
with different length. This makes our strategy a multi-dimensional asynchronous motif
detection with variable motif length.
In urban planning, sophisticated simulation models are key tools to estimate future population growth for measuring the impact of planning decisions on urban developments and the environment. Simulated population projections usually result in large, macro-scale, multivariate geospatial data sets. Millions of records have to be processed, stored, and visualized to help planners explore and analyze complex population patterns. We introduce a database driven framework for visualizing geospatial multidimensional simulation data based on the output from UrbanSim, a software for the analysis and planning of urban developments. The designed framework is extendable and aims at integrating empirical-stochastic methods and urban simulation models with techniques developed for information visualization and cartography. First, we develop an empirical model for the estimation of residential building types based on demographic household characteristics. The predicted dwelling type information is important for the analysis of future material use, carbon footprint calculations, and for visualizing simultaneously the results of land usage, density, and other significant parameters in 3D space. Our model uses multinomial logistic regression to derive building types at different scales. The estimated regression coefficients are applied to UrbanSim output in order to predict residential building types. The simulation results and the estimated building types are managed in an object-relational geodatabase. From the database, density, building types, and significant demographic variables are visually encoded as scalable, georeferenced 3D geometries and displayed on top of aerial photographs in a Google Earth visual synthesis. The geodatabase can be accessed and the visualization parameters can be chosen through a web-based user interface. The geometries are encoded in KML, Google's markup language, as ready-to-visualize data sets. The goal is to enhance human cognition by displaying abstract representations of multidimensional data sets in a realistic context and thus to support decision making in planning processes.
Due to the steadily growing flood of data, the appropriate use of visualizations for efficient data analysis is as important today as it has never been before. In many application domains, the data flood is based on processes that can be represented by node-link diagrams. Within such a diagram, nodes may represent intermediate results (or products), system states (or snapshots), milestones or real (and possibly georeferenced) objects, while links (edges) can embody transition conditions, transformation processes or real physical connections. Inspired by the engineering sciences application domain and the research project “SinOptiKom: Cross-sectoral optimization of transformation processes in municipal infrastructures in rural areas”, a platform for the analysis of transformation processes has been researched and developed based on a geographic information system (GIS). Caused by the increased amount of available and interesting data, a particular challenge is the simultaneous visualization of several visible attributes within one single diagram instead of using multiple ones. Therefore, two approaches have been developed, which utilize the available space between nodes in a diagram to display additional information.
Motivated by the necessity of appropriate result communication with various stakeholders, a concept for a universal, dashboard-based analysis platform has been developed. This web-based approach is conceptually capable of displaying data from various data sources and has been supplemented by collaboration possibilities such as sharing, annotating and presenting features.
In order to demonstrate the applicability and usability of newly developed applications, visualizations or user interfaces, extensive evaluations with human users are often inevitable. To reduce the complexity and the effort for conducting an evaluation, the browser-based evaluation framework (BREF) has been designed and implemented. Through its universal and flexible character, virtually any visualization or interaction running in the browser can be evaluated with BREF without any additional application (except for a modern web browser) on the target device. BREF has already proved itself in a wide range of application areas during the development and has since grown into a comprehensive evaluation tool.
The visualization of numerical fluid flow datasets is essential to the engineering processes that motivate their computational simulation. To address the need for visual representations that convey meaningful relations and enable a deep understanding of flow structures, the discipline of Flow Visualization has produced many methods and schemes that are tailored to a variety of visualization tasks. The ever increasing complexity of modern flow simulations, however, puts an enormous demand on these methods. The study of vortex breakdown, for example, which is a highly transient and inherently three-dimensional flow pattern with substantial impact wherever it appears, has driven current techniques to their limits. In this thesis, we propose several novel visualization methods that significantly advance the state of the art in the visualization of complex flow structures. First, we propose a novel scheme for the construction of stream surfaces from the trajectories of particles embedded in a flow. These surfaces are extremely useful since they naturally exploit coherence between neighboring trajectories and are highly illustrative in nature. We overcome the limitations of existing stream surface algorithms that yield poor results in complex flows, and show how the resulting surfaces can be used a building blocks for advanced flow visualization techniques. Moreover, we present a visualization method that is based on moving section planes that travel through a dataset and sample the flow. By considering the changes to the flow topology on the plane as it moves, we obtain a method of visualizing topological structures in three-dimensional flows that are not accessible by conventional topological methods. On the same algorithmic basis, we construct an algorithm for the tracking of critical points in such flows, thereby enabling the treatment of time-dependent datasets. Last, we address some problems with the recently introduced Lagrangian techniques. While conceptually elegant and generally applicable, they suffer from an enormous computational cost that we significantly use by developing an adaptive approximation algorithm. This allows the application of such methods on very large and complex numerical simulations. Throughout this thesis, we will be concerned with flow visualization aspect of general practical significance but we will particularly emphasize the remarkably challenging visualization of the vortex breakdown phenomenon.
In urban planning, both measuring and communicating sustainability are among the most recent concerns. Therefore, the primary emphasis of this thesis concerns establishing metrics and visualization techniques in order to deal with indicators of sustainability.
First, this thesis provides a novel approach for measuring and monitoring two indicators of sustainability - urban sprawl and carbon footprints – at the urban neighborhood scale. By designating different sectors of relevant carbon emissions as well as different household categories, this thesis provides detailed information about carbon emissions in order to estimate impacts of daily consumption decisions and travel behavior by household type. Regarding urban sprawl, a novel gridcell-based indicator model is established, based on different dimensions of urban sprawl.
Second, this thesis presents a three-step-based visualization method, addressing predefined requirements for geovisualizations and visualizing those indicator results, introduced above. This surface-visualization combines advantages from both common GIS representation and three-dimensional representation techniques within the field of urban planning, and is assisted by a web-based graphical user interface which allows for accessing the results by the public.
In addition, by focusing on local neighborhoods, this thesis provides an alternative approach in measuring and visualizing both indicators by utilizing a Neighborhood Relation Diagram (NRD), based on weighted Voronoi diagrams. Thus, the user is able to a) utilize original census data, b) compare direct impacts of indicator results on the neighboring cells, and c) compare both indicators of sustainability visually.
Nowadays, the increasing demand for ever more customizable products has emphasized the need for more flexible and fast-changing manufacturing systems. In this environment, simulation has become a strategic tool for the design, development, and implementation of such systems. Simulation represents a relatively low-cost and risk-free alternative for testing the impact and effectiveness of changes in different aspects of manufacturing systems.
Systems that deal with this kind of data for its use in decision making processes are known as Simulation-Based Decision Support Systems (SB-DSS). Although most SB-DSS provide a powerful variety of tools for the automatic and semi-automatic analysis of simulations, visual and interactive alternatives for the manual exploration of the results are still open to further development.
The work in this dissertation is focused on enhancing decision makers’ analysis capabilities by making simulation data more accessible through the incorporation of visualization and analysis techniques. To demonstrate how this goal can be achieved, two systems were developed. The first system, viPhos – standing for visualization of Phos: Greek for light –, is a system that supports lighting design in factory layout planning. viPhos combines simulation, analysis, and visualization tools and techniques to facilitate the global and local (overall factory or single workstations, respectively) interactive exploration and comparison of lighting design alternatives.
The second system, STRAD - standing for Spatio-Temporal Radar -, is a web-based systems that considers the spatio/attribute-temporal analysis of event data. Since decision making processes in manufacturing also involve the monitoring of the systems over time, STRAD enables the multilevel exploration of event data (e.g., simulated or historical registers of the status of machines or results of quality control processes).
A set of four case studies and one proof of concept prepared for both systems demonstrate the suitability of the visualization and analysis strategies adopted for supporting decision making processes in diverse application domains. The results of these case studies indicate that both, the systems as well as the techniques included in the systems can be generalized and extended to support the analysis of different tasks and scenarios.
Due to remarkable technological advances in the last three decades the capacity of computer systems has improved tremendously. Considering Moore's law, the number of transistors on integrated circuits has doubled approximately every two years and the trend is continuing. Likewise, developments in storage density, network bandwidth, and compute capacity show similar patterns. As a consequence, the amount of data that can be processed by today's systems has increased by orders of magnitude. At the same time, however, the resolution of screens has hardly increased by a factor of ten. Thus, there is a gap between the amount of data that can be processed and the amount of data that can be visualized. Large high-resolution displays offer a way to deal with this gap and provide a significantly increased screen area by combining the images of multiple smaller display devices. The main objective of this dissertation is the development of new visualization and interaction techniques for large high-resolution displays.
This dissertation focuses on the visualization of urban microclimate data sets,
which describe the atmospheric impact of individual urban features. The application
and adaptation of visualization and analysis concepts to enhance the
insight into observational data sets used this specialized area are explored, motivated
through application problems encountered during active involvement
in urban microclimate research at the Arizona State University in Tempe, Arizona.
Besides two smaller projects dealing with the analysis of thermographs
recorded with a hand-held device and visualization techniques used for building
performance simulation results, the main focus of the work described in
this document is the development of a prototypic tool for the visualization
and analysis of mobile transect measurements. This observation technique involves
a sensor platform mounted to a vehicle, which is then used to traverse
a heterogeneous neighborhood to investigate the relationships between urban
form and microclimate. The resulting data sets are among the most complex
modes of in-situ observations due to their spatio-temporal dependence, their
multivariate nature, but also due to the various error sources associated with
moving platform observations.
The prototype enables urban climate researchers to preprocess their data,
to explore a single transect in detail, and to aggregate observations from multiple
traverses conducted over diverse routes for a visual delineation of climatic
microenvironments. Extending traditional analysis methods, the suggested visualization
tool provides techniques to relate the measured attributes to each
other and to the surrounding land cover structure. In addition to that, an
improved method for sensor lag correction is described, which shows the potential
to increase the spatial resolution of measurements conducted with slow
air temperature sensors.
In summary, the interdisciplinary approach followed in this thesis triggers
contributions to geospatial visualization and visual analytics, as well as to urban
climatology. The solutions developed in the course of this dissertation are
meant to support domain experts in their research tasks, providing means to
gain a qualitative overview over their specific data sets and to detect patterns,
which can then be further analyzed using domain-specific tools and methods.