Regular physical activity is essential to maintain or even improve an individual’s health. There exist various guidelines on how much individuals should do. Therefore, it is important to monitor performed physical activities during people’s daily routine in order to tell how far they meet professional recommendations. This thesis follows the goal to develop a mobile, personalized physical activity monitoring system applicable for everyday life scenarios. From the mentioned recommendations, this thesis concentrates on monitoring aerobic physical activity. Two main objectives are defined in this context. On the one hand, the goal is to estimate the intensity of performed activities: To distinguish activities of light, moderate or vigorous effort. On the other hand, to give a more detailed description of an individual’s daily routine, the goal is to recognize basic aerobic activities (such as walk, run or cycle) and basic postures (lie, sit and stand).
With recent progress in wearable sensing and computing the technological tools largely exist nowadays to create the envisioned physical activity monitoring system. Therefore, the focus of this thesis is on the development of new approaches for physical activity recognition and intensity estimation, which extend the applicability of such systems. In order to make physical activity monitoring feasible in everyday life scenarios, the thesis deals with questions such as 1) how to handle a wide range of e.g.
everyday, household or sport activities and 2) how to handle various potential users. Moreover, this thesis deals with the realistic scenario where either the currently performed activity or the current user is unknown during the development and training
phase of activity monitoring applications. To answer these questions, this thesis proposes and developes novel algorithms, models and evaluation techniques, and performs thorough experiments to prove their validity.
The contributions of this thesis are both of theoretical and of practical value. Addressing the challenge of creating robust activity monitoring systems for everyday life the concept of other activities is introduced, various models are proposed and validated. Another key challenge is that complex activity recognition tasks exceed the potential of existing classification algorithms. Therefore, this thesis introduces a confidence-based extension of the well known AdaBoost.M1 algorithm, called ConfAdaBoost.M1. Thorough experiments show its significant performance improvement compared to commonly used boosting methods. A further major theoretical contribution is the introduction and validation of a new general concept for the personalization of physical activity recognition applications, and the development of a novel algorithm (called Dependent Experts) based on this concept. A major contribution of practical value is the introduction of a new evaluation technique (called leave-one-activity-out) to simulate when performing previously unknown activities in a physical activity monitoring system. Furthermore, the creation and benchmarking of publicly available physical activity monitoring datasets within this thesis are directly benefiting the research community. Finally, the thesis deals with issues related to the implementation of the proposed methods, in order to realize the envisioned mobile system and integrate it into a full healthcare application for aerobic activity monitoring and support in daily life.
The recognition of day-to-day activities is still a very challenging and important research topic. During recent years, a lot of research has gone into designing and realizing smart environ- ments in different application areas such as health care, maintenance, sports or smart homes. As a result, a large amount of sensor modalities were developed, different types of activity and context recognition services were implemented and the resulting systems were benchmarked using state-of-the-art evaluation techniques. However, so far hardly any of these approaches have found their way into the market and consequently into the homes of real end-users on a large scale. The reason for this is, that almost all systems have one or more of the following characteristics in common: expensive high-end or prototype sensors are used which are not af- fordable or reliable enough for mainstream applications; many systems are deployed in highly instrumented environments or so-called "living labs", which are far from real-life scenarios and are often evaluated only in research labs; almost all systems are based on complex system con- figurations and/or extensive training data sets, which means that a large amount of data must be collected in order to install the system. Furthermore, many systems rely on a user and/or environment dependent training, which makes it even more difficult to install them on a large scale. Besides, a standardized integration procedure for the deployment of services in existing environments and smart homes has still not been defined. As a matter of fact, service providers use their own closed systems, which are not compatible with other systems, services or sensors. It is clear, that these points make it nearly impossible to deploy activity recognition systems in a real daily-life environment, to make them affordable for real users and to deploy them in hundreds or thousands of different homes.
This thesis works towards the solution of the above mentioned problems. Activity and context recognition systems designed for large-scale deployment and real-life scenarios are intro- duced. Systems are based on low-cost, reliable sensors and can be set up, configured and trained with little effort, even by technical laymen. It is because of these characteristics that we call our approach "minimally invasive". As a consequence, large amounts of training data, that are usu- ally required by many state-of-the-art approaches, are not necessary. Furthermore, all systems were integrated unobtrusively in real-world/similar to real-world environments and were evalu- ated under real-life, as well as similar to real-life conditions. The thesis addresses the following topics: First, a sub-room level indoor positioning system is introduced. The system is based on low-cost ceiling cameras and a simple computer vision tracking approach. The problem of user identification is solved by correlating modes of locomotion patterns derived from the trajectory of unidentified objects and on-body motion sensors. Afterwards, the issue of recognizing how and what mainstream household devices have been used for is considered. Based on a low-cost microphone, the water consumption of water-taps can be approximated by analyzing plumbing noise. Besides that, operating modes of mainstream electronic devices were recognized by using rule-based classifiers, electric current features and power measurement sensors. As a next step, the difficulty of spotting subtle, barely distinguishable hand activities and the resulting object interactions, within a data set containing a large amount of background data, is addressed. The problem is solved by introducing an on-body core system which is configured by simple, one-time physical measurements and minimal data collections. The lack of large training sets is compensated by fusing the system with activity and context recognition systems, that are able to reduce the search space observed. Amongst other systems, previously introduced approaches and ideas are revisited in this section. An in-depth evaluation shows the impact of each fusion procedure on the performance and run-time of the system. The approaches introduced are able to provide significantly better results than a state-of-the-art inertial system using large amounts of training data. The idea of using unobtrusive sensors has also been applied to the field of behavior analysis. Integrated smartphone sensors are used to detect behavioral changes of in- dividuals due to medium-term stress periods. Behavioral parameters related to location traces, social interactions and phone usage were analyzed to detect significant behavioral changes of individuals during stressless and stressful time periods. Finally, as a closing part of the the- sis, a standardization approach related to the integration of ambient intelligence systems (as introduced in this thesis) in real-life and large-scale scenarios is shown.
The automatic analysis and retrieval of technical line drawings is hindered by many challenges such as: the large amount of contextual clutter around the symbols within the drawings, degradation, transformations on the symbols in drawings, large databases of drawings
and large alphabets of symbols. The core tasks required for the analysis of technical line
drawings are: symbol recognition, spotting and retrieval. The current systems for performing these tasks have poor performance due to the mentioned challenges. This dissertation
presents a number of methods that address these challenges. These methods achieve both
accurate and efficient symbol spotting and retrieval in technical line drawings, and perform
significantly better than state-of-the-art methods on the same problems. An overview of
the key contributions of this dissertation is given in the following.
First, this dissertation presents a geometric matching-based method for symbol recognition
and spotting. The method performs recognition in the presence of large amounts of contextual clutter, and provides precise localization of the recognized symbols. On standard
databases such as GREC-2005 and GREC-2011, the method achieves up to 10% higher
recall and up to 28% higher precision than state-of-the-art methods on the spotting task,
and achieves up to 7% higher recognition accuracy on the isolated recognition task. The
method is based on a geometric matching approach, which is flexible enough to incorporate
improvements on the matching strategy, feature types and information on the features. The
method also includes an adaptive preprocessing algorithm that deals with a wide variety
of noise types.
In order to improve the performance of the spotting method when dealing with degraded
drawings, two novel methods are presented in this dissertation. Both methods are based on
combining geometric matching with machine learning techniques. The geometric matching
is used to automatically generate training data that contain information on how well the
features of the queries are matched in both the true and the false matches found by the
spotting method. The first method learns the feature weights of the different query symbols
by linear discriminant analysis (LDA). The weighted query features are used in the spotting
method and result in 27% higher average precision than the original method, with a speedup
factor of 2. The second method uses SVM classification as a post-spotting step to distinguish
the true from the false matches in the spotting method. The use of the classification step
further improves the average precision of the spotting method by 20.6%.
This dissertation also presents methods for content analysis of line drawings. First, a
method for accurate and consistent detection (95.8%) of regions of interest (ROIs) is presented. The method is based on statistical feature grouping. The ROI-finding method is
identified as an important part of a symbol retrieval system: the better the detected ROIs,the higher the performance of a retrieval system. The ROI-finding method is also used to
improve the performance of the geometric-based spotting system.
Second, a symbol clustering method for building a compact and accurate representation of
a large database of technical drawings is presented. This method uses the output from the
ROI-finding method as input, and uses geometric matching as a similarity measure. The
method achieves high accuracy (90.1% recall, 94.3% precision) in forming clusters of symbols. The representatives of the clusters (34 symbols) are used as key entries to a symbol
index, which is identified as the outcome of an off-line stage of a symbol retrieval system.
Finally, an efficient and high performing large scale symbol retrieval system is presented
in this dissertation. The system follows the bag of visual words (BoVW) model, but with
using methods that are suitable to line drawings. The system uses the symbol index to
represent a database of drawings. During the on-line query retrieval stage, the query is
analyzed by the ROI-finding method, matched with the key entries of the symbol index via
geometric matching, and finally, a spatial verification step is performed on the retrieved
matches. The system achieves a query lookup time that is independent of the size of the
database, and is instead dependent on the size of the symbol index. The system achieves up
to 10% higher recall and up to 28% higher precision than state-of-the-art spotting systems
on similar databases.
Overall, these contributions are major advancements in the research of graphics recognition.
The hope is that, such contributions provide the basis for the development of reliable and
accurate performing applications for browsing, querying or classification of line drawings
for the benefit of end users.
Data usage control is a concept that extends access control to also protect data after it
has been released. Usage control enforcement relies on available information about the
distribution of data in the monitored system. In this thesis we introduce an information
engine V8 of the Chromium browser to evaluate the feasibility of the chosen approach.
Fluid extraction is a typical chemical process where two types of fluids are mixed together. The high complexity of this process which involves droplet coalescence, breakup, mass transfer, and counter-current flow often makes design difficult. The industrial design of these processes is still based on expensive mini-plant and pilot plant experiments. Therefore, there is a strong need for research into the stimulation of fluid-fluid interaction processes using computational fluid dynamics (CFD).
Previous multi-phase fluid simulations have focused on the development of models that couple mass and momentum using the Navier-Stokes equation. Recent population balance models (PBM) have proved to be important methods for analyzing droplet breakage and collisions. A combination of CFD and PBM facilitates the simulation of flow property by solving coupling equations, and the calculation of the droplet size and numbers. In our study, we successfully coupled an Euler-Euler CFD model with the breakup and coalescence models proposed by Luo and Svendsen (59).
The simulation output of extraction columns provides a mathematical understand- ing of how fluids are mixed inside a mixing device. This mixing process shows that the dispersed phase of a flow generates large blobs and bubbles. Current mathemati- cal simulation results often fail to provide an intuitive representation of how well two different types of fluid interact, so intuitive and physically plausible visualization tech- niques are in high demand to help chemical engineers to explore and analyze bubble column simulation data. In chapter 3, we present the visualization tools we developed for extraction column data.
Fluid interfaces and free surfaces are topics of growing interest in the field of multi- phase computational fluid dynamics. However, the analysis of the flow field relative to the material interface shape and topology is a challenging task. In chapter 5, we present a technique that facilitates the visualization and analysis of complex material interface behaviors over time. To achieve this, we track the surface parameterization of time-varying material interfaces and identify locations where there are interactions between the material interfaces and fluid particles. Splatting and surface visualization techniques produce an intuitive representation of the derived interface stability. Our results demonstrate that the interaction of a flow field with a material interface can be understood using appropriate extraction and visualization techniques, and that our techniques can help the analysis of mixing and material interface consistency.
In addition to texture-based methods for surface analysis, the interface of two- phase fluid can be considered as an implicit function of the density or volume fraction values. High-level visualization techniques such as topology-based methods can re- veal the hidden structure underlying simple simulation data, which will enhance and advance our understanding of multi-fluid simulation data. Recent feature-based vi- sualization approaches have explored the possibility of using Reeb graphs to analyze scalar field topologies(19, 107). In chapter 6, we present a novel interpolation scheme for interpolating point-based volume fraction data and we further explore the implicit fluid interface using a topology-based method.
Data integration aims at providing uniform access to heterogeneous data, managed by distributed source systems. Data sources can range from legacy systems, databases, and enterprise applications to web-scale data management systems. The materialized approach to data integration, extracts data from the sources, transforms and consolidates the data, and loads it into an integration system, where it is persistently stored and can be queried and analyzed.
To support materialized data integration, so called Extract-Transform-Load (ETL) systems have been built and are widely used to populate data warehouses today. While ETL is considered state-of-the-art in enterprise data warehousing, a new paradigm known as MapReduce has recently gained popularity for web-scale data transformations, such as web indexing or page rank computation.
The input data of both, ETL and MapReduce programs keeps changing over time, while business transactions are processed or the web is crawled, for instance. Hence, the results of ETL and MapReduce programs get stale and need to be recomputed from time to time. Recurrent computations over changing input data can be performed in two ways. The result may either be recomputed from scratch or recomputed in an incremental fashion. The idea behind the latter approach is to update the existing result in response to incremental changes in the input data. This is typically more efficient than the full recomputation approach, because reprocessing unchanged portions of the input data can often be avoided.
Incremental recomputation techniques have been studied by the database research community mainly in the context of the maintenance of materialized views and have been adopted by all major commercial database systems today. However, neither today's ETL tools nor MapReduce support incremental recomputation techniques. The situation of ETL and MapReduce programmers nowadays is thus much comparable to the situation of database programmers in the early 1990s. This thesis makes an effort to transfer incremental recomputation techniques into the ETL and MapReduce environments. This poses interesting research challenges, because these environments differ fundamentally from the relational world with regard to query and programming models, change data capture, transactional guarantees and consistency models. However, as this thesis will show, incremental recomputations are feasible in ETL and MapReduce and may lead to considerable efficiency improvements.
As a Software Product Line (SPL) evolves with increasing number of features and feature values, the feature correlations become extremely intricate, and the specifications of these correlations tend to be either incomplete or inconsistent with their realizations, causing misconfigurations in practice. In order to guide product configuration processes, we present a solution framework to recover complex feature correlations from existing product configurations. These correlations are further pruned automatically and validated by domain experts. During implementation, we use association mining techniques to automatically extract strong association rules as potential feature correlations. This approach is evaluated using a large-scale industrial SPL in the embedded system domain, and finally we identify a large number of complex feature correlations.
In recent years, recommender systems have been widely used for a variety of different kinds of items such as books, movies, and music. However, current recommendation approaches have often been criticized to suffer from overspecialization thus not enough considering a user’s diverse topics of interest. In this thesis we present a novel approach to extracting contextualized user profiles which enable recommendations taking into account a user’s full range of interests. The method applies algorithms from the domain of topic detection and tracking to automatically identify diverse user interests and to represent them with descriptive labels. That way manual annotations of interest topics by the users, e. g., from a predefined domain taxonomy, are no longer required. The approach has been tested in two scenarios: First, we implemented a content-based recommender system for an Enterprise 2.0 resource sharing platform where the contextualized user interest profiles have been used to generate recommendations with a high degree of inter-topic diversity. In an effort to harness the collective intelligence of the users, the resources in the system were described by making use of user-generated metadata. The evaluation experiments show that our approach is likely to capture a multitude of diverse interest topics per user. The labels extracted are specific for these topics and can be used to retrieve relevant on-topic resources. Second, a slightly adapted variation of the algorithm has been used to target music recommendations based on the user’s current mood. In this scenario music artists are described by using freely available Semantic Web data from the Linked Open Data cloud thus not requiring expensive metadata annotations by experts. The evaluation experiments conducted show that many users have a multitude of different preferred music styles. However a correlation between these music styles and music mood categories could not be observed. An integration of our proposed user profiles with existing user model ontologies seems promising for enabling context-sensitive recommendations.
Most innovation in the automotive industry is driven by embedded systems. They make usage of dynamic adaption to environmental changes or component/subsystem failures for remaining safe. Following this evolution, fault tree analysis techniques have been extended with concept for dynamic adaptation but resulting techniques like state event fault tree analysis, are not widely used in practice.
In this report we present the results of a controlled experiment that analyze these two techniques (State Events Fault Trees and Faul trees combined with markov chains) with regard to their applicability and efficiency in modeling dynamic behavior of dynamic embedded systems.
The experiment was conducted with students of the TU Kaiserslautern to modeli different safety aspects of an ambient assisted living system.
The main results of the experiment show that SEFTs where more easy and effective to use.
Most of the evolution in ambient assisted living is due to embedded
systems that dynamically adapt themself to react to environmental
changes or component/subsystem failures to maintain a certain level of
safety. Following this evolution fault tree analysis techniques have been
extended with concept for dynamic adaptation but resulting techniques
such as dynamic fault trees or state event fault trees analysis are not
widely used as expected.
In this report we describe a controlled experiment to analyze these two
techniques with regard to their applicability and efficiency in modeling
dynamic behavior of ambient assisted living systems.
Results of the experiment show that Dynamic Fault Trees are easier and more effective
to use, although they produce better results (models) with State Events Fault Trees.