Bøger udgivet af Technische Universität Chemnitz

The property of human motion perception is used in this dissertation to infer human activity from data using artificial neural networks. One of the main aims of this thesis is to discover which modalities, namely RGB images, optical flow and human keypoints, are best suited for HAR in omnidirectional data. Since these modalities are not yet available for omnidirectional cameras, they are synthetically generated with a 3D indoor simulation with the result of a large-scale dataset, called OmniFlow. Due to the lack of omnidirectional optical flow data, the OmniFlow dataset is validated using Test-Time Augmentation. Compared to the baseline, which contains Recurrent All-Pairs Field Transforms trained on the FlyingChairs and FlyingThings3D datasets, it was found that only about 1000 images need to be used for fine-tuning to obtain a very low End-point Error. For an evaluation on activity-level, two state-of-the-art convolutional neural networks (CNNs), namely the Temporal Segment Network (TSN) for the modalities RGB images and optical flow and the PoseC3D for the modality human keypoints, were used. Both CNNs were trained and validated on OmniFlow and on the real-world dataset OmniLab. For both networks, TSN and PoseC3D, three hyperparameters were varied and the top-1, top-5 and mean accuracies were reported. In addition, confusion matrices indicating the class-wise accuracy of the 15 activity classes have been given for the modalities RGB images, optical flow and human keypoints.

Vibration converters based on piezoelectric materials are currently becoming increasingly important for powering low-power wireless sensor nodes and wearable electronic devices. Piezoelectric materials generate variable electrical charges under mechanical stress, requiring an energy management interface to meet load requirements. Resonant interfaces like Parallel Synchronized Switch Harvesting on Inductor (P-SSHI) are highly efficient and robust to energy sources and loads variations. Nevertheless, SSHI circuits require synchronous switch control for efficient energy transfer. At irregular excitation, SSHI circuits may not perform optimally because the resonant frequency of the circuit is typically tuned to match the frequency of the energy source, which in the case of footsteps can be irregular and unpredictable. In addition, the circuit may also be susceptible to noise and interference from irregular excitations, which can further affect its performance. The aim is to design a self-powered energy management solution that can operate autonomously even at low frequencies and for irregular chock excitations, while at the same time allowing higher energy flow to the energy storage device and maintaining high levels of energy efficiency. To evaluate the performance of the proposed circuit, a piezoelectric shoe insole is designed and used for testing with different storage capacitance values and loads as a proof of the circuit's adaptability to various loading conditions.

Data gathering is a constant in human history with ever increasing amounts in quantity and dimensionality. To get a feel for the data, make it interpretable, or find underlying laws it is necessary to fit a function to the finite and possibly noisy data. In this thesis we focus on a method achieving this, namely least squares approximation. Its discovery dates back to around 1800 and it has since then proven to be an indispensable tool which is efficient and has the capability to achieve optimal error when used right.Crucial for the least squares method are the ansatz functions and the sampling points. To discuss them, we gather tools from probability theory, frame subsampling, and L2-Marcinkiewicz-Zygmund inequalities. With that we give results in the worst-case or minmax setting, when a set of points is sought for approximating a class of functions, which we model as a generic reproducing kernel Hilbert space. Further, we give error bounds in the statistical learning setting for approximating individual functions from possibly noisy samples. Here, we include the covariate-shift setting as a subfield of transfer learning. In a natural way a parameter choice question arises for balancing over- and underfitting effect. We tackle this by using the cross-validation score, for which we show a fast way of computing as well as prove the goodness thereof.

In the dissertation "Through-Silicon Vias in SiGe BiCMOS and Interposer Technologies for sub-THz Applications", a through-silicon via (TSV) technology module is developed targeting high frequency packaging applications in the millimeter-wave and sub-THz frequency range. Based on a high-performance 130 nm SiGe BiCMOS technology, TSVs are embedded in a via-middle integration approach. The TSV process module consists of the TSV integration, a carrier wafer handling and a wafer backside process to realize a redistribution layer. A process flow was developed to realize TSV structures with various geometries to ensure a high TSV design flexibility. The TSV process module can be applied for SiGe BiCMOS as well as for interposer technologies.High-performance TSV interconnections have been simulated to optimize their electrical properties. TSVs were electrically characterized and 3D transitions with low insertion loss

In this thesis, we study the following three types of periodic string patterns and some of their variants.Firstly, we consider maximal d-repetitions. These are substrings that are at least 2+d times as long as their minimum period.Secondly, we consider 3-cadences. These are arithmetic subsequence of three equal characters.Lastly, we consider maximal pairs. These are pairs of identical substrings.Maximal d-repetitions and maximal pairs of uncompressed strings are already well-researched. However, no non-trivial upper bound for distinct occurrences of these patterns that take the compressed size of the underlying strings into account were known prior to this research.We provide upper bounds for several variants of these two patterns that depend on the compressed size of the string, the logarithm of the string's length, the highest allowed power and d.These results also lead to upper bounds and new insights for the compacted directed acyclic word graph and the run-length encoded Burrows-Wheeler transform.We prove that cadences with three elements can be efficiently counted in uncompressed strings and can even be efficiently detected on grammar-compressed binary strings. We also show that even slightly more difficult variants of this problem are already NP-hard on compressed strings.Along the way, we extend the underlying geometry of the convolution from rectangles to arbitrary polygons. We also prove that this non-rectangular convolution can still be efficiently computed.

Security is a relatively new topic in the automotive industry. In the former days, the only security defense methods were the engine immobilizer and the anti-theft alarm system. The rising connection of vehicles to external networks made it necessary to extend the security effort by introducing security development processes. These processes include, among others, risk analysis and treatment steps. In parallel, the development of ISO/SAE 21434 and UNECE No. R155 started. The long development cycles in the automotive industry made it necessary to align the development processes' early designs with the standards' draft releases.This work aims to design a new consistent, complete, and efficient security development process aligned with the normative references. The resulting development process design aligns with the overall development methodology of the underlying, evaluated development process. Use cases serve as a basis for evaluating improvements and the method designs.This work concentrates on the left leg of the V-Model. Nevertheless, future work targets extensions for a holistic development approach for safety and security.