J. Computer Applications
Refine
Year of publication
Document Type
- Doctoral Thesis (13)
- Report (2)
- Article (1)
Has Fulltext
- yes (16)
Keywords
- Ambient Intelligence (2)
- Elektrotechnik (2)
- Szenario (2)
- deep learning (2)
- Adjoint method (1)
- Azidifizierung (1)
- B-Spline (1)
- BMP1 MAP-Kinase (1)
- Botrytis cinerea (1)
- Botrytis fabae (1)
Faculty / Organisational entity
- Kaiserslautern - Fachbereich Informatik (9)
- Kaiserslautern - Fachbereich Elektrotechnik und Informationstechnik (3)
- Kaiserslautern - Fachbereich Bauingenieurwesen (1)
- Kaiserslautern - Fachbereich Biologie (1)
- Kaiserslautern - Fachbereich Mathematik (1)
- Kaiserslautern - Fachbereich Wirtschaftswissenschaften (1)
Data-driven and Sparse-to-Dense Concepts in Scene Flow Estimation for Automotive Applications
(2022)
Highly assisted driving and autonomous vehicles require a detailed and accurate perception of the environment. This includes the perception of the 3D geometry of the scene and the 3D motion of other road users. The estimation of both based on images is known as the scene flow problem in computer vision. This thesis deals with a solution to the scene flow problem that is suitable for application in autonomous vehicles. This application imposes strict requirements on accuracy, robustness, and speed. Previous work was lagging behind in at least one of these metrics. To work towards the fulfillment of those requirements, the sparse-to-dense concept for scene flow estimation is introduced in this thesis. The idea can be summarized as follows: First, scene flow is estimated for some points of the scene for which this can be done comparatively easily and reliably. Then, an interpolation is performed to obtain a dense estimate for the entire scene. Because of the separation into two steps, each part can be optimized individually. In a series of experiments, it is shown that the proposed methods achieve competitive results and are preferable to previous techniques in some aspects. As a second contribution, individual components in the sparse-to-dense pipeline are replaced by deep learning modules. These are a highly localized and highly accurate feature descriptor to represent pixels for dense matching, and a network for robust and generic sparse-to-dense interpolation. Compared to end-to-end architectures, the advantage of deep modules is that they can be trained more effciently with data from different domains. The recombination approach applies a similar concept as the sparse-to-dense approach by solving and combining less diffcult, auxiliary sub-problems. 3D geometry and 2D motion are estimated separately, the individual results are combined, and then also interpolated into a dense scene flow. As a final contribution, the thesis proposes a set of monolithic end-to-end networks for scene flow estimation.
Robuste Optimierung wird zur Entscheidungsunterstützung eines komplexen Beschaffungs- und Transportmodells genutzt, um die Risikoeinstellung der Entscheidenden abzubilden und gleichzeitig ein robustes Ergebnis zu erzielen. Die Modellierung des Problems ist umfassend dargestellt und Ergebnisse der nicht-deterministischen Planung bei verschiedenen Parametern und Risikoeinstellungen gegenübergestellt. Die Datenunsicherheit wird an einem Praxisfall erläutert und Methoden und -empfehlungen zum Umgang mit dieser dargestellt.
The development of machine learning algorithms and novel sensing modalities has boosted the exploration of human activity recognition(HAR) in recent years. In this work, we explored field-based sensing solutions and different machine learning models for HAR tasks to address the shortcomings of existing HAR sensing solutions, like the weak robustness of RF-based solution, environment-dependency of the optic-based solution, etc., aiming to supply a competitive and alternative sensing approach for HAR tasks.
Field, in physics, describes a region in which each point will be affected by force. Field sensing is potentially a low-cost, low-power, non-intrusive, privacy-respecting HAR solution that is ideal for long-term, wearable activity recording. By directly/indirectly monitoring the field strength or other field variation caused variables, some unsolved HAR problems could be addressed when other sensing solutions fail. An example is the social distance monitoring problem, where the most widely adopted approach is based on the Bluetooth signal strength measurement. However, the signal is so subtle that any object surrounding the signal emitter will cause signal attenuation. To guarantee the accuracy of social distance monitoring, we developed an induced magnetic field-based social distance monitoring system with an accuracy of a sub-ten centimetre. Moreover, the system is robust and resistant to environmental variations. Like Bluetooth, other RF-wave-based sensing modalities also face the multi-path effect caused by refraction. Thus their signal is unreliable for positioning applications where higher accuracy and robustness are needed. Besides the magnetic field, we also explored a natural static passive electric field, the field between the human body and surroundings, namely the human body capacitance(HBC). HBC is a physiological parameter describing the charge distribution difference between the body and the surroundings and is seldomly explored before. We developed a few wearable, low-cost, low power consumption hardware platforms, either based on an oscillating unit or discrete components composed sensing front end followed by a high resolution analog-to-digital module, to
monitor the variation of the parameter regarding the body movement and environmental variations. Compared with the inertial sensors, the HBC could deliver full-body movement perceiving, meaning that the movement of the legs could be perceived by a wrist-worn HBC sensing unit, which is far beyond the
sensing ability of an inertial sensing unit.
To summarize, we introduced two competitive field sensing modalities for HAR tasks, the magnetic field sensing for position-related services and the passive electric field sensing for full-body action and environmental variation sensing. Both of which were still in an infant stage and not fully explored in the community. The advantages of the two field sensing modalities were demonstrated with a series of position-related and motion-related experiments.
In this thesis we develop a shape optimization framework for isogeometric analysis in the optimize first–discretize then setting. For the discretization we use
isogeometric analysis (iga) to solve the state equation, and search optimal designs in a space of admissible b-spline or nurbs combinations. Thus a quite
general class of functions for representing optimal shapes is available. For the
gradient-descent method, the shape derivatives indicate both stopping criteria and search directions and are determined isogeometrically. The numerical treatment requires solvers for partial differential equations and optimization methods, which introduces numerical errors. The tight connection between iga and geometry representation offers new ways of refining the geometry and analysis discretization by the same means. Therefore, our main concern is to develop the optimize first framework for isogeometric shape optimization as ground work for both implementation and an error analysis. Numerical examples show that this ansatz is practical and case studies indicate that it allows local refinement.
This work presents a visual analytics-driven workflow for an interpretable and understandable machine learning model. The model is driven by a reverse
engineering task in automotive assembly processes. The model aims
to predict the assembly parameters leading to the given displacement field
on the geometries surface. The derived model can work on both measurement
and simulation data. The proposed approach is driven by the scientific
goals from visual analytics and interpretable artificial intelligence alike. First, a concept for systematic uncertainty monitoring, an object-oriented, virtual reference scheme (OOVRS), is developed. Afterward, the prediction task is solved via a regressive machine learning model using adversarial neural networks.
A profound model parameter study is conducted and assisted with an interactive visual analytics pipeline. Further, the effects of the learned
variance in displacement fields are analyzed in detail. Therefore a visual analytics pipeline is developed, resulting in a sensitivity benchmarking tool. This allows the testing of various segmentation approaches to lower the machine learning input dimensions. The effects of the assembly parameters are
investigated in domain space to find a suitable segmentation of the training
data set’s geometry. Therefore, a sensitivity matrix visualization is developed. Further, it is shown how this concept could directly compare results
from various segmentation methods, e.g., topological segmentation, concerning the assembly parameters and their impact on the displacement field variance. The resulting databases are still of substantial size for complex simulations with large and high-dimensional parameter spaces. Finally, the applicability of video compression techniques towards compressing visualization image databases is studied.
In the past, information and knowledge dissemination was relegated to the
brick-and-mortar classrooms, newspapers, radio, and television. As these
processes were simple and centralized, the models behind them were well
understood and so were the empirical methods for optimizing them. In today’s
world, the internet and social media has become a powerful tool for information
and knowledge dissemination: Wikipedia gets more than 1 million edits per day,
Stack Overflow has more than 17 million questions, 25% of US population visits
Yahoo! News for articles and discussions, Twitter has more than 60 million
active monthly users, and Duolingo has 25 million users learning languages
online. These developments have introduced a paradigm shift in the process of
dissemination. Not only has the nature of the task moved from being centralized
to decentralized, but the developments have also blurred the boundary between
the creator and the consumer of the content, i.e., information and knowledge.
These changes have made it necessary to develop new models, which are better
suited to understanding and analysing the dissemination, and to develop new
methods to optimize them.
At a broad level, we can view the participation of users in the process of
dissemination as falling in one of two settings: collaborative or competitive.
In the collaborative setting, the participants work together in crafting
knowledge online, e.g., by asking questions and contributing answers, or by
discussing news or opinion pieces. In contrast, as competitors, they vie for
the attention of their followers on social media. This thesis investigates both
these settings.
The first part of the thesis focuses on the understanding and analysis of
content being created online collaboratively. To this end, I propose models for
understanding the complexity of the content of collaborative online discussions
by looking exclusively at the signals of agreement and disagreement expressed
by the crowd. This leads to a formal notion of complexity of opinions and
online discussions. Next, I turn my attention to the participants of the crowd,
i.e., the creators and consumers themselves, and propose an intuitive model for
both, the evolution of their expertise and the value of the content they
collaboratively contribute and learn from on online Q&A based forums. The
second part of the thesis explores the competitive setting. It provides methods
to help the creators gain more attention from their followers on social media.
In particular, I consider the problem of controlling the timing of the posts of
users with the aim of maximizing the attention that their posts receive under
the idealized setting of full-knowledge of timing of posts of others. To solve
it, I develop a general reinforcement learning based method which is shown to
have good performance on the when-to-post problem and which can be employed in
many other settings as well, e.g., determining the reviewing times for spaced
repetition which lead to optimal learning. The last part of the thesis looks at
methods for relaxing the idealized assumption of full knowledge. This basic
question of determining the visibility of one’s posts on the followers’ feeds
becomes difficult to answer on the internet when constantly observing the feeds
of all the followers becomes unscalable. I explore the links of this problem to
the well-studied problem of web-crawling to update a search engine’s index and
provide algorithms with performance guarantees for feed observation policies
which minimize the error in the estimate of visibility of one’s posts.
Dieses Szenario ist eine Erweiterung eines Teilszenarios von Human Centered Manufacturing. Dabei geht es um die Montage der Energieelektrik für industrielle Anlagen. Im Jahr 2015 enthält die Ausrüstung eines Elektromonteurs bei der Verdrahtung von Schaltschränken u.a. einen Schutzhelm mit integrierter Farbkamera, integriertem Mikrofon und einem Lautsprecher im Ohrbereich sowie einen automatisch gesteuerten Laserpointer. Auf der Baustelle sind keine Pläne mehr erforderlich. Der Monteur benötigt keinen Plan während der Montage.
Botrytis cinerea, der Erreger der Graufäule, infiziert hunderte verschiedene Pflanzenspezies und verursacht weltweit enorme landwirtschaftliche Verluste. Dabei tötet er das Wirtsgewebe sehr schnell mithilfe lytischer Enzyme und Nekrose-induzierender Metaboliten und Proteine ab. Das Signal-Mucin Msb2 ist in B. cinerea, wie in anderen pathogenen Pilzen, wichtig für die Oberflächenerkennung, Differenzierung von Appressorien und die Penetration des Pflanzengewebes. Msb2 agiert oberhalb der BMP1 Pathogenitäts-MAPK-Kaskade. In dieser Studie konnte eine direkte Interaktion zwischen Msb2 und BMP1, sowie zwischen den beiden Sensorproteinen Msb2 und Sho1 nachgewiesen werden. Dennoch führte die Deletion von sho1 lediglich zu geringfügigen Defekten im Wachstum, der Hyphendifferenzierung und der Bildung von Infektionsstrukturen. Sho1 zeigte nur einen geringen Einfluss auf die Aktivierung von BMP1. Das Fehlen von Sho1 verursachte keine Virulenzdefekte, während der Doppel-KO von msb2 und sho1 zu einer stärkeren Reduzierung der Läsionsausbreitung im Vergleich zu msb2 Mutanten führte. Es wurden mehrere keimungsregulierte, BMP1 abhängige Gene deletiert und die Mutanten phänotypisch charakterisiert. Keines der Gene für lytische Enzyme oder putative Effektorproteine beeinflusste die Virulenz. Mutanten, denen das für ein Ankyrin-repeat Protein codierende arp1 Gen fehlt, zeigten eine gestörte Oberflächenerkennung, gravierende Wachstumsdefekte und reduzierte Virulenz.
B. cinerea VELVET-Mutanten sind in der lichtabhängigen Differenzierung und der Ausbreitung nekrotischer Läsionen beeinträchtigt. In dieser Arbeit ermöglichte die Charakterisierung mehrerer Mutanten ein besseres Verständnis der molekularen Vorgänge, aufgrund derer der VELVET-Komplex die Entwicklung und Pathogenese in B. cinerea reguliert. Quantitative Vergleiche der in planta Transkriptome und Sekretome führten zur Identifizierung eines für drei VELVET-Mutanten gemeinsamen Sets an herunterregulierten Genen, welche für CAZymes, Proteasen und Virulenz-assoziierte Proteine codieren. Die meisten dieser Gene wurden zusätzlich im Wildtyp während der Infektion verstärkt exprimiert, was zusätzliche Hinweise auf deren Relevanz im Infektionsprozess lieferte. Die drastisch verringerte Expression von Genen für Proteasen konnte mit niedrigerer Proteaseaktivität und der unvollständigen Mazeration des Gewebes an der Infektionsstelle in Verbindung gebracht werden. Der neu etablierte quantitative Sekretom-Vergleich des Wildtyps und der VELVET-Mutanten mithilfe 15N-markierter Proteine korrelierte eindeutig mit den Transkriptomdaten sekretierter Proteine. Damit wurde gezeigt, dass die Abundanz der Proteine maßgeblich von deren mRNA-Level abhängt. Die Unfähigkeit zur Ansäuerung des Wirtsgewebes ist einer der interessantesten phänotypischen Aspekte der VELVET-Mutanten. Während Citrat die dominierende von B. cinerea ausgeschiedene Säure ist, sekretierten VELVET-Mutanten deutlich weniger Citrat. Weder für Oxalat noch für Gluconat konnte eine wichtige Rolle während der Infektion bestätigt werden. Die Läsionsausbreitung der Mutanten wurde sowohl durch Zugabe von Vollmedium, als auch durch künstlich konstant niedrig eingestellte pH-Werte an den Infektionsstellen gefördert, während die Einstellung auf neutrale pH-Werte die Expansion beim B. cinerea Wildtyp stark beeinträchtigte. Damit ist die Ansäuerung in Tomatenblättern ein wichtiger Virulenzmechanismus, der möglicherweise essentiell für die Aktivität der sekretierten Proteine ist.
Überraschenderweise scheint eine Ansäuerung des Gewebes für die erfolgreiche Infektion der Ackerbohne Vicia faba nicht notwendig zu sein. Weder B. cinerea noch der am nächsten verwandte Botrytis fabae, welcher sich als Spezialist auf V. faba aggressiver verhält, zeigten während der erfolgreichen Infektion eine Ansäuerung des Ackerbohnenblattgewebes. B. fabae ist auf wenige Wirtspflanzen der Fabaceae begrenzt. Die Grundlagen der Wirtsspezifität sind bisher unklar. Vergleichende Transkriptom- und Sekretom-Analysen ergaben Hinweise für die molekularen Ursachen der unterschiedlichen Wirtsspektren von B. cinerea und B. fabae. In dieser Arbeit konnte die schlechte Infektion durch B. fabae auf Tomatenblättern mit einer deutlich niedrigeren Proteaseaktivität in Verbindung gebracht werden, während artifiziell konstant niedrige pH-Werte die Läsionsausbreitung kaum förderten. Im Gegensatz zur Infektion von Tomatenblättern wurden jedoch auf V. faba insgesamt deutlich niedrigere Proteaseaktivitäten in den Sekretomen beider Spezies gemessen. Diese Daten weisen darauf hin, dass die beiden Spezies nicht nur generell unterschiedliche Infektionsstrategien anwenden, sondern dass die Virulenzmechanismen zusätzlich vom infizierten Pflanzengewebe abhängig sind.
The neural networks have been extensively used for tasks based on image sensors. These models have, in the past decade, consistently performed better than other machine learning methods on tasks of computer vision. It is understood that methods for transfer learning from neural networks trained on large datasets can reduce the total data requirement while training new neural network models. These methods tend not to perform well when the data recording sensor or the recording environment is unique from the existing large datasets. The machine learning literature provides various methods for prior-information inclusion in a learning model. Such methods employ methods like designing biases into the data representation vectors, enforcing priors or physical constraints on the models. Including such information into neural networks for the image frames and image-sequence classification is hard because of the very high dimensional neural network mapping function and little information about the relation between the neural network parameters. In this thesis, we introduce methods for evaluating the statistically learned data representation and combining these information descriptors. We have introduced methods for including information into neural networks. In a series of experiments, we have demonstrated methods for adding the existing model or task information to neural networks. This is done by 1) Adding architectural constraints based on the physical shape information of the input data, 2) including weight priors on neural networks by training them to mimic statistical and physical properties of the data (hand shapes), and 3) by including the knowledge about the classes involved in the classification tasks to modify the neural network outputs. These methods are demonstrated, and their positive influence on the hand shape and hand gesture classification tasks are reported. This thesis also proposes methods for combination of statistical and physical models with parametrized learning models and show improved performances with constant data size. Eventually, these proposals are tied together to develop an in-car hand-shape and hand-gesture classifier based on a Time of Flight sensor.
The goal of this work is to develop statistical natural language models and processing techniques
based on Recurrent Neural Networks (RNN), especially the recently introduced Long Short-
Term Memory (LSTM). Due to their adapting and predicting abilities, these methods are more
robust, and easier to train than traditional methods, i.e., words list and rule-based models. They
improve the output of recognition systems and make them more accessible to users for browsing
and reading. These techniques are required, especially for historical books which might take
years of effort and huge costs to manually transcribe them.
The contributions of this thesis are several new methods which have high-performance computing and accuracy. First, an error model for improving recognition results is designed. As
a second contribution, a hyphenation model for difficult transcription for alignment purposes
is suggested. Third, a dehyphenation model is used to classify the hyphens in noisy transcription. The fourth contribution is using LSTM networks for normalizing historical orthography.
A size normalization alignment is implemented to equal the size of strings, before the training
phase. Using the LSTM networks as a language model to improve the recognition results is
the fifth contribution. Finally, the sixth contribution is a combination of Weighted Finite-State
Transducers (WFSTs), and LSTM applied on multiple recognition systems. These contributions
will be elaborated in more detail.
Context-dependent confusion rules is a new technique to build an error model for Optical
Character Recognition (OCR) corrections. The rules are extracted from the OCR confusions
which appear in the recognition outputs and are translated into edit operations, e.g., insertions,
deletions, and substitutions using the Levenshtein edit distance algorithm. The edit operations
are extracted in a form of rules with respect to the context of the incorrect string to build an
error model using WFSTs. The context-dependent rules assist the language model to find the
best candidate corrections. They avoid the calculations that occur in searching the language
model and they also make the language model able to correct incorrect words by using context-
dependent confusion rules. The context-dependent error model is applied on the university of
Washington (UWIII) dataset and the Nastaleeq script in Urdu dataset. It improves the OCR
results from an error rate of 1.14% to an error rate of 0.68%. It performs better than the
state-of-the-art single rule-based which returns an error rate of 1.0%.
This thesis describes a new, simple, fast, and accurate system for generating correspondences
between real scanned historical books and their transcriptions. The alignment has many challenges, first, the transcriptions might have different modifications, and layout variations than the
original book. Second, the recognition of the historical books have misrecognition, and segmentation errors, which make the alignment more difficult especially the line breaks, and pages will
not have the same correspondences. Adapted WFSTs are designed to represent the transcription. The WFSTs process Fraktur ligatures and adapt the transcription with a hyphenations
model that allows the alignment with respect to the varieties of the hyphenated words in the line
breaks of the OCR documents. In this work, several approaches are implemented to be used for
the alignment such as: text-segments, page-wise, and book-wise approaches. The approaches
are evaluated on German calligraphic (Fraktur) script historical documents dataset from “Wan-
derungen durch die Mark Brandenburg” volumes (1862-1889). The text-segmentation approach
returns an error rate of 2.33% without using a hyphenation model and an error rate of 2.0%
using a hyphenation model. Dehyphenation methods are presented to remove the hyphen from
the transcription. They provide the transcription in a readable and reflowable format to be used
for alignment purposes. We consider the task as classification problem and classify the hyphens
from the given patterns as hyphens for line breaks, combined words, or noise. The methods are
applied on clean and noisy transcription for different languages. The Decision Trees classifier
returns better performance on UWIII dataset and returns an accuracy of 98%. It returns 97%
on Fraktur script.
A new method for normalizing historical OCRed text using LSTM is implemented for different texts, ranging from Early New High German 14th - 16th centuries to modern forms in New
High German applied on the Luther bible. It performed better than the rule-based word-list
approaches. It provides a transcription for various purposes such as part-of-speech tagging and
n-grams. Also two new techniques are presented for aligning the OCR results and normalize the
size by using adding Character-Epsilons or Appending-Epsilons. They allow deletion and insertion in the appropriate position in the string. In normalizing historical wordforms to modern
wordforms, the accuracy of LSTM on seen data is around 94%, while the state-of-the-art combined rule-based method returns 93%. On unseen data, LSTM returns 88% and the combined
rule-based method returns 76%. In normalizing modern wordforms to historical wordforms, the
LSTM delivers the best performance and returns 93.4% on seen data and 89.17% on unknown
data.
In this thesis, a deep investigation has been done on constructing high-performance language
modeling for improving the recognition systems. A new method to construct a language model
using LSTM is designed to correct OCR results. The method is applied on UWIII and Urdu
script. The LSTM approach outperforms the state-of-the-art, especially for unseen tokens
during training. On the UWIII dataset, the LSTM returns reduction in OCR error rates from
1.14% to 0.48%. On the Nastaleeq script in Urdu dataset, the LSTM reduces the error rate
from 6.9% to 1.58%.
Finally, the integration of multiple recognition outputs can give higher performance than a
single recognition system. Therefore, a new method for combining the results of OCR systems is
explored using WFSTs and LSTM. It uses multiple OCR outputs and votes for the best output
to improve the OCR results. It performs better than the ISRI tool, Pairwise of Multiple Sequence and it helps to improve the OCR results. The purpose is to provide correct transcription
so that it can be used for digitizing books, linguistics purposes, N-grams, and part-of-speech
tagging. The method consists of two alignment steps. First, two recognition systems are aligned
using WFSTs. The transducers are designed to be more flexible and compatible with the different symbols in line and page breaks to avoid the segmentation and misrecognition errors.
The LSTM model then is used to vote the best candidate correction of the two systems and
improve the incorrect tokens which are produced during the first alignment. The approaches
are evaluated on OCRs output from the English UWIII and historical German Fraktur dataset
which are obtained from state-of-the-art OCR systems. The Experiments show that the error
rate of ISRI-Voting is 1.45%, the error rate of the Pairwise of Multiple Sequence is 1.32%, the
error rate of the Line-to-Page alignment is 1.26% and the error rate of the LSTM approach has
the best performance with 0.40%.
The purpose of this thesis is to contribute methods providing correct transcriptions corresponding to the original book. This is considered to be the first step towards an accurate and
more effective use of the documents in digital libraries.