Research Projects

The goal of this project is to develop and analyze methods for analyzing textual information. This should be realized by using semi-supervised learning methods, which use labeled as well as unlabeled data. In particular, existing methods which are already applied for pattern recognition should be adapted such that those can also be applied for textual data. For a practical analysis comparisons to SVM and k-NN classifier using a boosting algorithm should be performed, the influence of the amount of labeled/unlabeled data and the convergence should be analyzed. Moreover, a fair comparative study between batch and on-line methods is performed.

Die klinische Rechtsmedizin gewann in den letzten Jahren aufgrund einer Sensibilisierung der Öffentlichkeit gegenüber häuslicher und sexueller Gewalt, Gewalt gegenüber Kindern und Verdachtsfällen von medizinischen Behandlungsfehlern stark an Bedeutung. Die forensische Untersuchung von Lebenden ist bis heute jedoch auf eine äussere Besichtigung des Körpers beschränkt.

Das neue Ludwig-Boltzmann-Institut (LBI) für klinisch-forensische Bildgebung hat zum Ziel, Verfahren zur Erfassung von inneren Verletzungsbefunden als Grundlage für forensische Gutachten zu entwickeln. Mittels Computertomographie (CT) und Magnetresonanztomographie (MRT), welche in der Klinik etabliert sind, können zusätzliche, objektiv nachweisbare innere Verletzungsbefunde erhoben werden, die eine verbesserte Einschätzung der ausgeübten Gewalt gegen die untersuchte Person ermöglichen. Die Methoden sind jedoch auf klinische Diagnostik ausgerichtet, während forensisch wichtige Befunde nicht oder nicht optimal dargestellt werden.

Das Institut fuer Maschinelles Sehen und Darstellen kooperiert mit dem LBI zur Entwicklung neuer Methoden der Bildverarbeitung und Computergrafik zum Zwecke der Bildgebung.

The research aim of the project is to provide a system infrastructure to support teams of users in the on-site monitoring of events and analysis of natural resources. The project will introduce the innovative concept of event-driven campaigns using handheld devices, potentially supported by an unmanned aerial vehicle (UAV). Event-driven campaigns provide users the capacity to analyse and predict environmental changes on-site, supporting the process of taking appropriate countermeasures to avoid environmental degradation. During these campaigns, users will be able to setup and retrieve data from mobile sensorstations, the UAV and external sources (such as permanent sensor networks) in order to generate dense information on a small area. The whole sensor network system will gather and store sensor data, and process simulations based on physical process models. Hence, a shared information system fusing heterogeneous data sources will be provided that supports teams of stakeholders to monitor environmental processes on-site, complementing remote monitoring and management. To enrich the data sets from a specific location, additional remotely controlled cameras will be deployed, mounted on sensorstations and below the UAV. Users will be able to analyse the environment using mobile phones and handheld computers, supported by advanced user interface techniques.

The project will improve monitoring and management for environmental scientists, institutions, service providers, engineering companies and municipalities through its strong integration of handhelds and sensor networks. The project will progress well beyond the current state in the art, by dealing with short-term events and detailed analysis of small sites. The analysis of such events is hardly supported by current methods, but has a large impact on environmental degradation. Furthermore, information is made available to citizens by providing mechanisms to access top-level environmental data. Within the project, cutting edge inter-disciplinary research will be performed to develop user-centered solutions. When the data is integrated with analytical tools in a shared information space it will also aid a wide range of managers and planners pursuing more environmentally sensitive solutions to engineering problems. To aid the process, the research is steered by considerable end-user involvement in all its phases.

The generation of realistic 3D models of whole cities has become a vibrant and highly competitive market through the recent activities of, most notably, Goggle Earth and Microsoft Virtual Earth. While the first generation of these systems only delivered high-quality zoomable images of the ground, the current trend is heavily geared towards 3D – that is, users can access three-dimensional height- fields of the terrain, and even 3D models of individual buildings. Simple building models, basically extruded polygons with different types of roofs, can be generated today from aerial images completely automatically. This is a solved problem. Far from solved, however, is the problem of generating automatically detailed buildings with façades. Input data for this problem are registered range maps obtained by stereo matching and sequences of highly overlapping thus redundant images (taken from a car driving in the road) where each pixel has not only a color but also a depth, a z-value. Although range maps can be directly rendered in principle, the data size is huge and, more importantly, the pixels have no semantics: A priori there is no difference between a pixel on the floor, on the wall, or on a door. But these shape semantics are required by all downstream applications using the city model. Shape grammars, on the other hand, have recently become (again) a popular method in research for representing 3D buildings. Their great advantage is that they allow to parameterize buildings, which can be used for populating virtual cities with believable architectural buildings, e.g., for 3D games. The goal of the CITYFIT project is, given highly redundant input imagery and range maps from an arbitrary building in Graz, to synthesize a shape grammar that, when evaluated, creates a clean, CAD- quality reconstruction of that building that fits the original data very closely and makes the semantics of all major architectural features explicit. These shape semantics can even be transferred back to inform the original data, so each of these “semantically enriched” data points can tell whether it belongs to ground, wall, or door.

The use of augmented reality in medicine is an important field, especially in teaching and training of sensitive tasks. To support teaching and training of neonatal cranial sonography, an augmented reality simulator was developed. Physical models of a newborn and an ultrasound probe were tracked and their movements displayed in their virtual representation. The head of the newborn model was augmented with a 3D volume, reconstructed from ultrasound images of a real patient. Reconstructing a 3D volume from irregular source data takes a special focus on positioning the images and the subsequent interpolation. Moving the physical model towards each other, the according slices are generated in realtime.

VIPEM ist ein System zur hypothesengesteuerten Analyse multidimensionaler Datenräume im Gebiet der personalisierten Medizin. Ein multidimensionaler Datenraum, bestehend aus molekularen und klinischen Daten, wird unter gleichzeitiger Anwendung algorithmischer Verfahren und direkter Benutzerinteraktion gefiltert und hierarchisch strukturiert. Ein zentrales Forschungsproblem der personalisierten Medizin ist die Frage, wie die Verknüpfungen zwischen genetischen Variationen und Krankheiten, bzw. dem Ansprechen auf bestimmte Medikamente, gefunden werden können. Dazu gilt es, z.B. Gendaten mit klinischen Daten zu verknüpfen und in Folge spezifische Patientengruppen zu identifizieren. Die großen Datenmengen der molekularen Analyseverfahren (genetische Polymorphismen, Genexpressionsdaten, Proteomics) können nur mehr mit Methoden der Bioinformatik und Statistik bewältigt werden. Aber auch Standardmethoden der Statistik und der Bioinformatik versagen, wenn die Daten sehr inhomogen strukturiert sind dies ist bei den klinischen Daten der Fall und wenn Strukturen in den Daten durch Rauschen bzw. dominante Muster verdeckt werden. VIPEM soll mit Hilfe von Visualierungsmethoden die Struktur in den Datenräumen sichtbar machen und eine interaktive Navigation und Strukturierung sowohl der molekularen, als auch der klinischen Daten erlauben. VIPEM baut auf Grundlagenergebnissen in den Bereichen Informations-Visualisierung und multimodale Benutzerschittstellen auf. Durch eine enge Verknüpfung mehrerer gleichzeitig wirksamer Eingabekanäle und die sofortige Sichtbarkeit der Analyseschritte in der Visualisierung steht dem Experten ein Werkzeug zu interaktiven Erkundung von komplexen Datenräumen zur Verfügung. Als Eingabeparameter für Analysealgorithmen nutzt VIPEM hierbei die menschliche Fähigkeit, komplexe Muster und Zusammenhänge visuell bereits in Ansätzen zu erfassen, und erlaubt dadurch das Freilegen sonst verdeckter Strukturen. VIPEM fokussiert auf die hohe Nachfrage nach visualisierter Analytik im Bereich der Bioinformatik. Der innovative Zugang von VIPEM versteht sich als einmaliges Verkaufsargument, zumal sich mit VIPEM ein viel versprechendes Produkt abzeichnet, welches sicher innerhalb der nächsten zwei bis drei Jahre seinen Stellenwert als verwertbares Produkt am Markt behaupten könnte. Diese Forschungsarbeit wird als Teil des Projekts Caleydo durchgeführt.

Genoptikum is an interactive data exploration system for the visualization of and navigation in molecular and clinical data in the field of personalized medicine. Genoptikum addresses the essential but to date unsolved problem of how to identify connections between genetic variants and their corresponding diseases or the response to certain drugs and treatments, respectively. It is, therefore, necessary to connect gene data and clinical data in order to categorise specific subgroups of patients with certain disease features. The huge amount of data provided by molecular analytical methods (genetic polymorphisms, gene expression data, proteomics) can only be analysed by applying statistical methods and bioinformatics. However, even standard methods of statistics and bioinformatics fail when the data are inhomogeneous as is the case with clinical data and when data structures are obscured by noise and dominant patterns. Genoptikum should make the structure of the data spaces visible by using innovative methods of visualisation based on multiple high resolution displays in combination with data projection technologies. Genoptikum is bases on fundamental results in the fields of visualisation of information and multimodal user interfaces which enable an interactive navigation and structuring of both molecular and clinical data. Through a close link between several input channels, which are simultaneously active, and by immediate visualisation of the steps of the analysis, the expert is provides with a tool for the interactive exploration of complex data spaces. As input parameter for analysis algorithms Genoptikum makes use of the human visual capacity to grasp complex patterns to reveal hidden structures and correlations in large data spaces. This research is part of the project Caleydo.

Office space usually consists of private single-user workstations. Team work takes place on separate locations, usually supported by analogue media like printed paper. Digital data exchanges is accomplished through designated channels like e-mail or instant messengers.

Deskotheque is an ongoing project aiming to extend personal workspaces to enhance team work. It represents a flexible, interactive environment for team work, conference and meeting rooms. Unused surfaces in the room, such as empty wall space and table surfaces, can be turned into interactive, digital displays to be used for multi-user co-located teamwork.

The systematic creation of models of the real world to support the locational awareness on the Internet can be achieved if previously required massive manual labor gets replaced by automated procedures. A particular challenge exists in the automation of the extraction of the 4 classical map features buildings, circulation spaces (e.g. road networks), vegetation and water bodies, as well as their interaction. Decennia of research have been unable to automate the extraction of these features from classical aerial photography towards an economically viable result. However, we believe that we can succeed in the proposed project to develop automated procedures to create feature data for three reasons. First is the recent advent of digital aerial sensors producing highly redundant digital large format aerial photography. Redundancy will be obtained by using high forward and side overlaps, say at 80% and 60%, so that every point in the terrain is imaged at least 10 times, and any algorithm can rely on multiple analysis results that then can either reinforce or cancel one another. Second, the geometric redundancy gets augmented by a radiometric redundancy using 4 spectral bands, adding an infrared band to the classical red, green and blue color channels. Third, we will combine the classical "object reconstruction" approach available from stereo procedures, by new recognition methods. While classically a "car" on a street may have been seen via a "point cloud" and would have to get recognized simply by a representation of local height anomaly on an otherwise flat reference surface, recognition includes the use of stored images of cars in a data base to actually recognize a car as a human would do when inspecting an aerial image. The project is split up into five work packages which will focus on how reconstruction and recognition techniques can help each other and how additional information either from a previous mission or GIS can be integrated in the 3D modeling framework. One work package will address the assessment of the obtained quality, another will address project management and dissemination activities. Within the project we will develop an extensive library of combined recognition/reconstruction methods, and apply them to a range of test data sets. Test data will vary in geometric resolution (pixel size), overlaps, and types of terrain scenarios.

The trend in video surveillance is an ever increasing number of (digital) cameras for surveying complex scenarios (e.g. crowds). Currently available video surveillance systems cannot cope with this increased complexity, the detection rates are too low and the systems are not reliable enough. This hinders the broad use of automatic surveillance systems. AUTOVISTA proposes to use modern visual computing technologies to advance the state-of-the-art of video surveillance considerably. In order to cope with the increasing number of cameras, AUTOVISTA will (1) use novel on-line learning techniques to increase the detection rate and decrease the false alarm rate, while the camera adapts in an unsupervised manner to the surveyed scene. Besides an increased performance, this has the additional advantage that the installation and maintenance effort will be substantially decreased; (2) exploit novel visualization and interaction techniques to support the human operator. Furthermore two complementary visualization modes are proposed, blending smoothly between these allows the operator to maintain coherence. These techniques will enable a single operator to cope simultaneously with a large amount of cameras. AUTOVISTA will tackle the problem of increased people densities and highly cluttered scenes in a novel manner. Instead of relying on single person detection and tracking (which is not feasible for high people density scenarios), methods will be investigated to handle the crowd as a whole. AUTOVISTA will derive spatio-temporal crowd statistics, describe normal crowd behavior and use this for unusual event detection.