Part of
ISSN 2190-6629
The focus of this work is on human driving behavior in road traffic. Two aspects of it are covered, the prediction of it, including the identification of relevant influencing factors, as well as the behavior generation for autonomous vehicles.
The behavior prediction is based on a field study during which participants drove a measurement vehicle through inner-city traffic. Using the driven trajectories and lidar recordings complexity features to describe the surroundings at the intersection, the traffic there and the driving path are defined. The driving behavior is characterized by further features. Based on the complexity features regression models are trained to predict the behavior features. For that, linear regression, random forest and gradient boosting machine are utilized. Different complexity feature sets, including ones that are reduced with the help of an autoencoder, are used for prediction. The results show that the driving behavior can be predicted reliably. However, when using complexity feature sets with only few features the prediction performance is reduced.
In order to obtain a complexity score that is in line with human perception of complexity, an online study using videos of approaches to intersections was conducted. In pairwise comparisons participants were asked to identify the more complex situation. From that data complexity scores for the intersection passes included in the study are calculated. Several methods are used to assign these scores to the runs of the original field study. Behavior regression models are trained using these assigned complexity scores. The results show that behavior prediction with the complexity scores is possible, however, most variants require to also consider the turning direction as a second feature.
The behavior generation for decision-making at T-intersections is based on a discrete event system (DES). For it, several features are used to define events that describe the status of the decision-making process at the intersection. The events trigger the transitions between the states of the DES. All states are associated with either offensive or defensive driving behavior, which is implemented using the intelligent driver model. The algorithm is validated with a simulation framework. Using a generic map and several real maps, the decision-making model is simulated 14400 times while interacting with further cooperation vehicles. None of these runs resulted in a collision involving the vehicle running the algorithm and the times to pass the intersection can be explained by the numbers of cooperation vehicles and the intersection layouts. Further simulations are used to investigate the influence of limited visibility at the intersections on the model.
Umfang: XVII, 195 S.
Preis: 39.00 €
Buchdetails
In dieser Arbeit werden Ansätze zur verbesserten Signalanalyse mehrstimmiger Musikaufnahmen vorgestellt, die auf künstlichen neuronalen Netzen basieren. Diese Ansätze ermöglichen eine objektive Bewertung der Aufnahmequalität von Amateuraufnahmen, eine verbesserte zeitabhängige Detektion aktiver Musikinstrumente sowie eine bessere Separation von Ensemble-Aufnahmen mit unterschiedlichen Instrumenten.
In this work, improved signal analysis approaches for polyphonic music recordings, based on artificial neural networks, are presented. These approaches enable an objective estimation of the recording quality of amateur recordings, an improved time-dependent detection of active musical instruments, and an improved separation of ensemble recordings with different instruments.
Umfang: X, 184 S.
Preis: 39.00 €
Buchdetails
Deep learning is widely applied to sparse 3D data to perform challenging tasks, e.g., 3D object detection and semantic segmentation. However, the high performance of deep learning comes with high costs, including computational costs and the effort to capture and label data. This work investigates and improves the efficiency of deep learning for sparse 3D data to overcome the obstacles to the further development of this technology.
Umfang: XIII, 227 S.
Preis: 45.00 €
Buchdetails
The increasing use of automated laser welding processes causes high demands on process monitoring. This work demonstrates methods that use a camera mounted on the focussing optics to perform pre-, in-, and post-process monitoring of welding processes. The implementation uses machine learning methods. All algorithms consider the integration into industrial processes. These challenges include a small database, limited industrial manufacturing inference hardware, and user acceptance.
Umfang: XVI, 226 S.
Preis: 46.00 €
Buchdetails
Die vorliegende Arbeit stellt nun Verfahren zur Nutzung der Keramikglühkerze als Sensorelement vor. Das bedeutet, dass die Keramikglühkerze nicht nur ihrer konventionellen Aufgabe als Heizelement nachgeht, sondern auch als Sensor Informationen aus dem Brennraum (wie die Motordrehzahl) liefert. Dabei liegt das Hauptaugenmerk auf der Widerstandsänderung, insbesondere unter dem Einfluss von Temperaturänderungen im Brennraum, in unmittelbarer Umgebung der Keramikglühkerze.
This work presents methods for using the ceramic glow plug as a sensor element. This means that the ceramic glow plug will not only perform its conventional task as a heating element but will also provide information, such as motor speed, from the combustion chamber as a sensor. The main focus will be on the change in resistance, particularly under the influence of a temperature change, in the immediate vicinity of the ceramic glow plug.
Umfang: XII, 229 S.
Preis: 54.00 €
Buchdetails
Optical measurement methods are becoming increasingly important for high-precision production of components and quality assurance. The increasing demand can be met by modern imaging systems with advanced optics, such as light field cameras. This work explores their use in the deflectometric measurement of specular surfaces. It presents improvements in phase unwrapping and calibration techniques, enabling high surface reconstruction accuracies using only a single monocular light field camera.
Umfang: XI, 256 S.
Preis: 46.00 €
Buchdetails
In dieser Arbeit wird untersucht, wie überwacht trainierte künstliche neuronale Netze für die spektrale Entmischung eingesetzt werden können. Dazu wird zunächst eine geeignete Netzarchitektur ermittelt. Im weiteren Verlauf liegt der Schwerpunkt auf der Erzeugung geeigneter Trainingsdaten. Es werden modellbasierte Verfahren, die Trainingsdaten aus echten Reinspektren erzeugen, und datenbasierte Verfahren, die bereits vorhandene Trainingsdaten erweitern, vorgestellt und evaluiert.
In this work, artificial neural networks trained in a supervised manner for spectral unmixing are investigated. For this purpose, a suitable network architecture is determined first. After that, the focus lies on the generation of suitable training data. Model-based methods that generate training data from real pure spectra and data-based methods that augment existing training data are presented and evaluated.
Umfang: XII, 171 S.
Preis: 41.00 €
Buchdetails
This work presents model-based algorithmic approaches for interference-invariant time delay estimation, which are specifically suited for the estimation of small time delay differences with a necessary resolution well below the sampling time. Therefore, the methods can be applied particularly well for transit-time ultrasonic flow measurements, since the problem of interfering signals is especially prominent in this application.
Umfang: XIII, 188 S.
Preis: 44.00 €
Buchdetails
In dieser Arbeit werden spektral kodierte multispektrale Lichtfelder untersucht, wie sie von einer Lichtfeldkamera mit einem spektral kodierten Mikrolinsenarray aufgenommen werden. Für die Rekonstruktion der kodierten Lichtfelder werden zwei Methoden entwickelt, eine basierend auf den Prinzipien des Compressed Sensing sowie eine Deep Learning Methode. Anhand neuartiger synthetischer und realer Datensätze werden die vorgeschlagenen Rekonstruktionsansätze im Detail evaluiert.
In this work, spatio-spectrally coded multispectral light fields, as taken by a light field camera with a spectrally coded microlens array, are investigated. For the reconstruction of the coded light fields, two methods, one based on the principles of compressed sensing and one deep learning approach, are developed. Using novel synthetic as well as a real-world datasets, the proposed reconstruction approaches are evaluated in detail.
Umfang: VII, 214 S.
Preis: 45.00 €
Buchdetails
Im ersten Teil dieser Arbeit wird ein Framework für den Entwurf eines CNNs für FPGAs vorgestellt, das aus einem eigenen Vorverarbeitungsalgorithmus, einer Augmentierung, einem eigenen Quantisierungsschema und einer Verkleinerung des CNN besteht. Die Kombination von konventioneller Bildverarbeitung mit neuronalen Netzen wird im zweiten Teil anhand eines Beispiels aus der Robotik gezeigt, in dem ein bildbasierter Regler erfolgreich für einen Greifvorgang eines Roboters eingesetzt wird.
In the first part of this dissertation, a framework for the design of a CNN for FPGAs is presented, consisting of a preprocessing algorithm, an augmentation technique, a custom quantization scheme and a pruning step of the CNN. The combination of conventional image processing with neural networks is shown in the second part by an example from robotics, where an image-based visual servoing process is successfully conducted for a gripping process of a robot.
Umfang: XIV, 182 S.
Preis: 43.00 €
Buchdetails
