Refine
Document Type
- Bachelor Thesis (68) (remove)
Keywords
- Android <Systemplattform> (3)
- Routing (3)
- Akkreditierung (2)
- E-Government (2)
- Enterprise 2.0 (2)
- Smartphone (2)
- 3D-Kartierung (1)
- API (1)
- Ad-Hoc Routing (1)
- Agenten (1)
Institute
- Fachbereich 4 (68) (remove)
Im Vergleich zu herkömmlicher Computergrafik (perspektivische Projektion) bietet Raytracing entscheidende Vorteile, die hauptsächlich in der vergleichsweise hohen physikalischen Korrektheit der Methode begründet sind. Die Schwächen liegen hingegen im immensen Rechenaufwand.
Ein Raytracer ist vergleichsweise so rechenintensiv, weil für jeden Pixel mindestens ein Strahl verschickt werden muss. Dieser muss gegen alle Objekte im Raum geschnitten werden. Hinzu kommen noch die Strahlen, die entstehen, wenn Strahlen an Objekten reflektiert werden (Rekursion). Um diesen Rechenaufwand zu verkleinern und zusätzlich ein besseres Bild zu erzeugen, soll der adaptive Sampler den Raytracer unterstützen. Der adaptive Sampler soll während des Rendervorgangs den progressiven Fortschritt in der Bildgenerierung beobachten und Pixel von der weiteren Berechnung ausschließen, für die sich ein zusätzliches Verschießen von Strahlen nicht mehr lohnt.
Anders als der rein progressive Raytracer hört der adaptive Sampler mit dem Konvergieren des Bildes auf zu rechnen. Der adaptive Sampler soll so dafür sorgen, dass schneller ein besseres Bild erzeugt wird und somit die Performanz gesteigert wird. Insgesamt erwartet man sich vom adaptiven Sampler Vorteile bei der Berechnung von bestimmten Szenen. Unter anderem eine Verbesserung bei Szenen mit rein diffus beleuchteten Bildbereichen, sowie eine Verbesserung bei Szenen mit unterschiedlich rechenintensiven Bildbereichen. Ein normaler Raytracer kann nicht beurteilen, wie sinnvoll seine Schüsse sind. Er kann nur mehr Strahlen verschießen, in der Hoffnung, das Bild damit effektiv zu verbessern.
Es gibt jedoch viele Szenarien, bei denen eine linear steigende Schussanzahl pro Pixel keine gleichmäßige Verbesserung im Bild erzeugt. Das bedeutet, dass Bereiche im Bild schon gut aussehen, während andere noch sehr verrauscht sind. Man möchte nun Bildbereiche, die bereits konvergiert sind, in denen sich ein weiterer Beschuss also nicht mehr bemerkbar macht, ausschließen und die Rechenleistung dort nutzen, wo man sie noch braucht.
Wichtig dabei ist, dass Pixel nicht ungewollt zu früh von der Berechnung ausgeschlossen werden, die nicht weit genug konvergiert sind. Der adaptive Sampler soll so lange arbeiten, bis jeder Pixel dauerhaft keine Änderungen mehr vorweist. Das bedeutet, dass die Wahrscheinlichkeit für eine signifikante Farbänderung eines Pixels durch Verschießen eines Strahls (bei mehreren Lichtquellen in RenderGin mehrere Strahlen pro Pixel) klein genug ist. Es wird zwar intern keine Wahrscheinlichkeit berechnet, jedoch bekommt der Raytracer eine Art Gedächtnis: Er speichert die Veränderungen im beleuchteten Bild und deren Verlauf in eigenen Gedächtnisbildern. Das "Gedächtnis" für das alte Bild (Zustand des Bildes in der letzten Iteration über die Pixel) repräsentiert dabei das Kurzzeitgedächtnis. Es ist absolut genau. Das Langzeitgedächtnis wird von drei verschiedenen Bildern repräsentiert. Das erste gibt die Anzahl der verschossenen Strahlen pro Pixel an. Das zweite ist ein Wahrheitswertebild, das für jeden Pixel angibt, ob dieser noch in die Berechnung einbezogen werden soll. Das dritte Bild gibt an, wie oft jeder Pixel eine Farbänderung vollzogen hat, die geringer ist als der geforderte Maximalabstand eines Pixels zu sich selbst (vor und nach dem Verschießen eines weiteren Strahls).
Mit diesen drei Bildern ist es möglich, zusätzliche quantitative Informationen zu den qualitativen Informationen des Vergleichs vom neuen und alten Bild zu berücksichtigen.
In dieser Arbeit kläre ich die Frage, ob die gewünschten Effekte eintreten und ob bei Integration in die bestehende Struktur von RenderGin ein Performanzgewinn möglich ist. Die Umsetzung eines adaptiven Samplers ist als Plug-In in der Software RenderGin von Numenus GmbH geschehen. RenderGin ist ein echtzeitfähiger, progressiver Raytracer, der sich durch seine Performanz auszeichnet. Die Bildgenerierung geschieht allein auf der CPU, die Grafikkarte wird lediglich zur Anzeige des erzeugten Bildes benötigt.
Die Umsetzung und Programmierung des Plug-Ins ist in Microsoft Visual Studio 2010 geschehen unter Verwendung des RenderGin SDK der Numenus GmbH.
Due to the incorrect allocation management and the continuous increase of internet-capable devices, IPv4-adresses are nearly exhausted. For this reason and because of new requirements demanded by the technique of the internet the IPv6-protocol has been developed. The IPv6-protocol provides a larger adress space and will gradually replace the IPv4-protocol. However, the routing-protocols have to be adapted to the new internetprotocol. This paper will introduce and analyse the dynamic Routing-Information-Protocol RIPng which in itself is an inter-gateway protocol (IGP). The routers inside the network can exchange information about their different connections over this protocol. Furthermore, this paper will explain the basics of the IPv6-protocol, of the used protocol-algorithm as well as of the RIPv2-protocol. In the practical part of the paper the Counting-to-Infinity-Problem and some attributes of RIPng will be reviewed under the IPv6-protocol.
A Kinect device has the ability to record color and depth images simultaneously. This thesis is an attempt to use the depth image to manipulate lighting information and material properties in the color image. The presented method of lighting and material manipulation needs a light simulation of the lighting conditions at the time of recording the image. It is used to transform information from a new light simulation directly back into the color image. Since the simulations are performed on a three-dimensional model, a way is searched to generate a model out of single depth image. At the same time the text will react to the problems of the depth data acquisition of the Kinect sensor. An editor is designed to make lighting and material manipulations possible. To generate a light simulation, some simple, real-time capable rendering methods and lighting modells are proposed. They are used to insert new illumination, shadows and reflections into the scene. Simple environments with well defined lighting conditions are manipulated in experiments to show boundaries and possibilities of the device and the techniques being used.
Augmented Reality (AR) is getting more and more popular. To augment information into the field of vision of the user using HMDs, e.g. front shields of a car, glasses, displays of a smartphone or tablets are the main use of AR technology. It is necessary to get the position and orientation (pose) of the camera in space to augment correctly.
Nowadays, this is solved with artificial markers. These known markers are placed in the room and the system is taught to this set up. The next step is to get rid of these artificial markers. If we are calculating the pose without such markers we are talking about marker-less tracking. Instead of artificial markers we will use natural objects in the real world as reference points to calculate the pose. Thus, this approach can be used flexibly and dynamically. We are no longer dependent on artificial markers but we need much more knowledge about the scenery to find the pose. This is compensated by technical actions and/or the user himself. However, both solutions are neither comfortable nor efficient for the usage of such a system. This is why marker-less 3D tracking is still a big field of research.
This sets the starting point for the bachelor thesis. In this thesis an approach is proposed that needs only a quantity of 2D Feature from a given camera image and a quantity of 3D Feature of an object to find the initial Pose. With this approach, we got rid of the technical and user assistance. 2D and 3D Features can be detected in any way you like.
The main idea of this approach is to build six correspondences between these quantities. With those we are able to estimate the pose. Each 3D Feature is mapped with the estimated pose onto image coordinates, whereby the estimated pose can be evaluated. Each distance is measured between the mapped 3D Feature and the associated 2D Feature. Each correspondency is evaluated and the results are summed up to evaluate the whole pose. The lower this summed up value is, the better the pose. It has been shown to have a correct pose with a value around ten pixels.
Due to lots of possibilities to build six correspondences between the quantities, it is necessary to optimize the building process. For the optimization we will use a genetic algorithm.
During the test case the system worked quite reliably. The hit rate was around 90% with a runtime of approximately twelve minutes. Without optimization it can take easily some years.
Usability experts conduct user studies to identify existing usability problems. An established method is to record gaze behavior with an eye-tracker. These studies require a lot of effort to evaluate the results. Automated recognition of good and bad usability in recorded user data can support usability experts in eye tracking evaluation and reduce the effort. The objective of that bachelor thesis is to identify suitable eye-tracking metrics that correlate with the quality of usability. For this purpose, the central research question is answered: Which eye-tracking metrics correlate with the quality of a web form’s operation? To answer the research question, a quantitative A/B-user-study with eye-tracking was conducted and recorded the
gaze behavior of 30 subjects while filling out the web form. The web form was designed, that each web form page was available as a good and bad variant according to known usability guidelines. The results confirm a significant correlation between the eye-tracking-metric "number of visits to an
AOI" and the quality of the operation of a web form. The eye-tracking-metrics
"number of fixations within an AOI" and "duration of fixations within an AOI" also correlate with the quality of usability. No correlation could be confirmed for the "time of the first fixation within an AOI".
The goal of this Bachelor thesis was programming an existig six-legged robot, which should be able to explore any environment and create a map of it autonomous. A laser scanner is to be integrated for cognition of this environment. To build the map and locate the robot a suitable SLAM(Simultaneous Localization and Mapping) technique will be connected to the sensor data. The map is reported to be the robots base of path planning and obstancle avoiding, what will be developed in the scope of the bachelor thesis, too. Therefore both GMapping and Hector SLAM will be implemented and tested.
An exploration algorithm is described in this bachelor thesis for exploring the robots environment. The implementation on the robot takes place in the space of ROS(Robot Operating System) framework on a "Raspberry Pi" miniature PC.
Activity recognition with smartphones is possible by using its internal sensors without using any external sensor. First of all, previous works and their techniques will be regarded and from these works an own implementation for the activity recognition will be derived. Most of the previous works only use the accelerometer for the activity recognition task. For this reason, this bachelor thesis analyzes the benefit of further sensors, such as the magnetic field, the linear acceleration or the gyroscope. The activity recognition is performed by classification algorithms. Decision Tree, Naive Bayes and Support Vector machines will be used. Sensor data of subjects will be collected and saved by using an own developed application. This data is needed as training data for the classification algorithms.
The result is a model which represents the structure of the data. To validate the model, a test dataset will be used which is different from the training dataset. The results confirm previous works which indicated that the activity recognition task is possible by only using the accelerometer. Orientation, gyroscope and linear acceleration cannot be used for all problems of the activity recognition. Apart from that, the Decision Tree seems to be the best classification algorithm if the model has no training data of the current user.
Software integration is an engineering task where image-based approaches are still considered slow and error-prone. In this thesis, we apply image-based software integration to the domain of online Poker systems. That is, we implement the poker expert system PokerBot which can play online. We use the method of screen scraping to capture the screen information needed to interact with the Poker server. The consequent use of Template Matching leads to an efficient implementation. Substantial e ort was also addressed to the artificial intelligence aspects of PokerBot. The purpose of AI is here to mimic human playing style by translating a well-documented Pokerguide into formalized, executable language.
Autonomous systems such as robots already are part of our daily life. In contrast to these machines, humans an react appropriately to their counterparts. People can hear and interpret human speech, and interpret facial expressions of other people.
This thesis presents a system for automatic facial expression recognition with emotion mapping. The system is image-based and employs feature-based feature extraction. This thesis analyzes the common steps of an emotion recognition system and presents state-of-the-art methods. The approach presented is based on 2D features. These features are detected in the face. No neutral face is needed as reference. The system extracts two types of facial parameters. The first type consists of distances between the feature points. The second type comprises angles between lines connecting the feature points. Both types of parameters are implemented and tested. The parameters which provide the best results for expression recognition are used to compare the system with state-of-the-art approaches. A multiclass Support Vector Machine classifies the parameters.
The results are codes of Action Units of the Facial Action Coding System. These codes are mapped to a facial emotion. This thesis addresses the six basic emotions (happy, surprised, sad, fearful, angry, and disgusted) plus the neutral facial expression. The system presented is implemented in C++ and is provided with an interface to the Robot Operating System (ROS).
Only little information is available about the diffusion of cloud computing in German higher educational institutions. A better understanding of the state of the art in this field would support the modernization of the higher educational institutions in Germany and allow the development of more adequate cloud products and more appropriate business models for this niche. For this purpose, a literature research on Cloud Computing and IT-diffusion will be run and an empirical investigation with an online questionnaire addressed to higher educational institutions in Germany will be performed to illustrate the state of the art of Cloud Computing in German higher educational institutions as well as the threats and opportunities perceived by employees of higher educational institutions data centers connected to the usage of the cloud.
In addition to that, different experts from universities and businesses will be interviewed to complete the knowledge and information collected through the online questionnaire and during the research phase. The expected results will serve to create a recommendation for higher educational institutions in Germany about either they should migration to the cloud or not and introduce a list of guiding questions of critical issues to consider before using cloud-computing technologies.