Schenk, Michael (Ed.): 14. IFF-Wissenschaftstage 2011. Tagungsband : Digitales Engineering und virtuelle Techniken zum Planen, Testen und Betreiben technischer Systeme. Stuttgart: Fraunhofer Verlag, 2011, pp. 191-196

Verteilte Mixed Reality Systeme, d.h. VR/AR-Installationen, die sich an unterschiedlichen Standorten befinden und via Internet vernetzt sind, wurden bislang nur in akademischen Testszenarien eingesetzt. Sie konnten den Sprung in industrielle Nutzung jedoch nicht vollziehen. Im Szenario "VR-Meeting" des BMBF-Projektes AVILUS wurde auf Grundlage des Virtual und Augmented Reality Systems "instantreality" (www.instantreality.de) ein kollaboratives VR-System entwickelt, mit dem Entwicklungsteams an verteilten Standorten in einer virtuellen Konstruktionssitzung zusammenarbeiten. Dabei wurden zum einen Verfahren für eine sichere Datenverteilung realisiert, zum anderen wurden Methoden zur effizienten Protokollierung von verteilten immersiven Konstruktionssitzungen erforscht.

Wiedmer, Martin (Ed.) et al.: 10th IEEE International Symposium on Mixed and Augmented Reality 2011 : ISMAR. Science & Technology Proceedings, pp. 15-18

With "House of Olbrich" we present an iPhone Augmented Reality (AR) app that visualizes the compelling history of Darmstadt's unique Jugendstil (Art Nouveau) quarter with video-see through Augmented Reality. We propose methods for enabling high performance computer vision algorithms to deploy sophisticated AR visuals on current generation Smartphones by outsourcing resource intensive tasks to the cloud. This allows us to apply methods on 3D feature recognition for even complex tracking situations outdoors, where lightning conditions change and tracked objects are often occluded. By taking a snapshot of the building, the user learns about the architect, design and history of the building. Historical media, like old photographs and blueprints are superimposed on the building's front, depicting the moved history of the famous House of Olbrich, which was destroyed during World War II and has been only rudimentary restored. Augmented Reality technology allows tourists to jump back in time visually by using their Smartphones: Mixing Realities emphasizes the user's experience and leads his attention to the impressive historical architecture of the Art Nouveau. In addition, we ease interaction means by superimposing snapshots. Tourists may view and read information also in a relaxed position without the need to front-up their mobiles all the time.

Wiedmer, Martin (Ed.) et al.: 10th IEEE International Symposium on Mixed and Augmented Reality 2011. [USB Stick] : ISMAR. Science & Technology Proceedings, pp. 269-270

As large ships are never produced in masses, it often occurs that the construction process and production process overlap in time. Many shipbuilding companies have problems with discrepancies between the construction data and the real built ship. The assembly department often has to modify CAD data for a successful installation. We present an augmented reality system, where a user can visualize the construction data of pipes and modify these in the case of misalignment, collisions or any other conflicts. The modified pipe geometry can be stored and further used as input for CNC pipe bending machines. To guarantee an exactly orthogonal passage of the pipes through aligning bolt holes, we integrated an optical measurement tool into the pipe alignment process.

Computers & Graphics, Vol.35 (2011), 4, pp. 778-788

This paper focuses on the preparative process of natural feature map retrieval for a mobile camera-based tracking system. We cover the most important aspects of a general purpose tracking system including the acquisition of the scene's geometry, tracking initialization and fast and accurate frame-by-frame tracking. To this end, several state-of-the-art techniques - each targeted at one particular subproblem - are fused together, whereby their interplay and complementary benefits form the core of the system and the thread of our discussion. The choice of the individual sub-algorithms in our system reflects the scarcity of computational resources on mobile devices. In order to allow a more accurate, more robust and faster tracking during run-time, we therefore transfer the computational load into the preparative customization step wherever possible. From the viewpoint of the user, the preparative stage is kept very simple. It only involves recording the scene from various viewpoints and defining a transformation into a target coordinate frame via manual definition of only a few 3D to 3D point correspondences. Technically, the image sequence is used to (1) capture the scene's geometry by a SLAM-Method and subsequent refinement via constrained Bundle Adjustment, (2) to train a Randomized-Trees classifier for wide-baseline tracking initialization, and (3) to analyze the view-point dependent visibility of each feature. During run-time, robustness and performance of the frame-to-frame tracking are further increased by fusing inertial measurements within a combined pose estimation.

IEEE Computer Society: 3DIMPVT 2011 : International Conference on 3D Imaging, Modeling, Processing, Visualization and Transmission. Los Alamitos, Calif.: IEEE Computer Society Conference Publishing Services (CPS), 2011, pp. 140-147

This paper focuses on the preparative process of retrieving accurate feature maps for a camera-based tracking system. With this system it is possible to create ready-touse Augmented Reality applications with a very easy setup work-flow, which in practice only involves three steps: filming the object or environment from various viewpoints, defining a transformation between the reconstructed map and the target coordinate frame based on a small number of 3D-3D correspondences and, finally, initiating a feature learning and Bundle Adjustment step. Technically, the solution comprises several sub-algorithms. Given the image sequence provided by the user, a feature map is initially reconstructed and incrementally extended using a Simultaneous-Localization-and-Mapping (SLAM) approach. For the automatic initialization of the SLAM module, a method for detecting the amount of translation is proposed. Since the initially reconstructed map is defined in an arbitrary coordinate system, we present a method for optimally aligning the feature map to the target coordinated frame of the augmentation models based on 3D-3D correspondences defined by the user. As an initial estimate we solve for a rigid transformation with scaling, known as Absolute Orientation. For refinement of the alignment we present a modification of the well-known Bundle Adjustment, where we include these 3D-3D-correspondences as constraints. Compared to ordinary Bundle Adjustment we show that this leads to significantly more accurate reconstructions, since map deformations due to systematic errors such as small camera calibration errors or outliers are well compensated. This again results in a better alignment of the augmentations during run-time of the application, even in large-scale environments.

Avizzano, Carlo Alberto (Ed.) et al.: IEEE RO-MAN 2010 : 19th IEEE International Symposium on Robot and Human Interactive Communication. Proceedings [online]. New York: The Institute of Electrical and Electronics Engineers (IEEE), 2010, pp. 434-439

Augmented Reality (AR) is a technology which has become very popular in the last years. In this context also the idea of using of AR for training applications has become very important. AR offers a large potential for training only if the training is well focused to the skills that have to be trained and if the training protocol is well designed. On the other hand, the generation of the training content to be transferred via AR is a comprehensive problem that is addressed in this paper. Thus, this paper tries to describe the whole chain of implementations and general aspects involved in the creation of AR training applications, including examples for used multimodal devices. This chain starts with the capturing of expert actions to be hold in "digital representation of skill". The digital representation of skill is transferred to the training protocol that specifies the storyboard of the AR training session. The paper includes two different implementations of AR training systems and describes the general idea of informational abstraction from low level data up to interaction and from design to application.

Institute for Systems and Technologies of Information, Control and Communication (INSTICC): VISIGRAPP 2010. Proceedings : International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications. INSTICC Press, 2010, pp. 302-309

The most challenging algorithmical task for markerless Augmented Reality applications is the robust estimation of the camera pose. With a given 3D model of a scene the camera pose can be estimated via model-based camera tracking without the need to manipulate the scene with fiducial markers. Up to now, the bottleneck of model-based camera tracking is the availability of such a 3D model. Recently time-of-flight cameras were developed which acquire depth images in real time. With a sensor fusion approach combining the color data of a 2D color camera and the 3D measurements of a time-of-flight camera we acquire a textured 3D model of a scene. We propose a semi-manual reconstruction step in which the alignment of several submeshes with a mesh processing tool is supervised by the user to ensure a correct alignment. The evaluation of our approach shows its applicability for reconstructing a 3D model which is suitable for model-based camera tracking even for objects which are difficult to measure reliably with a time-of-flight camera due to their demanding surface characteristics.

Höllerer, Tobias (Ed.) et al.: 9th IEEE International Symposium on Mixed and Augmented Reality 2010 : ISMAR. Science & Technology Proceedings. Los Alamitos, Calif.: IEEE Computer Society, 2010, pp. 241-242

For many tasks like markerless model-based camera tracking it is essential that the 3D model of a scene accurately represents the real geometry of the scene. It is therefore very important to detect deviations between a 3D model and a scene. We present an innovative approach which is based on the insight that camera tracking can not only be used for Augmented Reality visualization but also to solve the correspondence problem between 3D measurements of a real scene and their corresponding positions in the 3D model. We combine a time-of-flight camera (which acquires depth images in real time) with a custom 2D camera (used for the camera tracking) and developed an analysis-by-synthesis approach to detect deviations between a scene and a 3D model of the scene.

Schenk, Michael (Ed.): 13. IFF-Wissenschaftstage 2010. Tagungsband : Digitales Engineering und virtuelle Techniken zum Planen, Testen und Betreiben technischer Systeme. Stuttgart: Fraunhofer Verlag, 2010, pp. 140-145

Um die globale Wettbewerbsfähigkeit nachhaltig zu sichern, die erhöhten technischen Anforderungen zu erfüllen und verkürzte Entwicklungszyklen bei gleichzeitig erhöhter Komplexität zu gewährleisten, ist der effiziente Einsatz von CAx- und virtuellen Technologien im Wertschöpfungsprozess für HDW unbedingte Voraussetzung. Dabei findet aufgrund von geringen Stückzahlen und hoher Individualität im U-Bootsbau eine enge Verzahnung von Fertigung und Entwicklung statt. Diese Verzahnung erfordert einen permanenten Abgleich zwischen digital konstruierten und real gefertigten Bauteilen. Um diese Anforderung zu unterstützen, werden im Rahmen des BMBF geförderten Projektes AVILUS Technologien entwickelt, die einerseits eine VR-gestützte Kooperation an verteilten Standorten unterstützten ("VRMeeting"), die andererseits aber mit Hilfe von Augmented Reality Technologien auch einen Abgleich von digitaler (Entwicklung) und realer Welt (Fertigung) unterstützen.

Latoschik, Marc Erich (Ed.) et al.: 2nd Workshop on Software Engineering and Architectures for Realtime Interactive Systems (SEARIS@VR2009) : IEEE Virtual Reality 2009 Workshop. Aachen: Shaker, 2009, pp. 51-54

In this paper we introduce an architecture for rapid development and assessment of advanced 3D visual track ing algorithms. We claim, that no universal tracking approach exists, that fulfills the requirements of all possible application seenarios at the same time. On the contrary, very specific and working solutions can be developed for given situations and uses. Therefore, software for visual tracking must be designed as a highly flexible system that can be quickly re-configured in order to enable the development of optimized solutions in terms of robustness, accuracy, frame rate and delay. To this purpose we designed an architecture that offers many functionalities, which can be combined together to build a new processing chain. The overall system offers numerous advantages, such as interactive programming, run-time access to data and parameters and easy interfacing with other libraries or applications.

Latoschik, Marc Erich (Ed.) et al.: 2nd Workshop on Software Engineering and Architectures for Realtime Interactive Systems (SEARIS@VR2009) : IEEE Virtual Reality 2009 Workshop. Aachen: Shaker, 2009, pp. 51-54

In this paper we introduce an architecture for rapid development and assessment of advanced 3D visual track ing algorithms. We claim, that no universal tracking approach exists, that fulfills the requirements of all possible application seenarios at the same time. On the contrary, very specific and working solutions can be developed for given situations and uses. Therefore, software for visual tracking must be designed as a highly flexible system that can be quickly re-configured in order to enable the development of optimized solutions in terms of robustness, accuracy, frame rate and delay. To this purpose we designed an architecture that offers many functionalities, which can be combined together to build a new processing chain. The overall system offers numerous advantages, such as interactive programming, run-time access to data and parameters and easy interfacing with other libraries or applications.

Institute for Systems and Technologies of Information, Control and Communication (INSTICC): VISIGRAPP 2009. Proceedings : International Joint Conference on Computer Vision and Computer Graphics Theory and Applications [CD-ROM]. INSTICC Press, 2009, GRAPP, pp. 387-392

This paper introduces novel techniques of interacting and controlling 3D content using multi-touch interaction principles for navigation and virtual camera control. Based on applications from GIS and for the architectural design review process, implementation and usage of these interaction techniques are illustrated. A comprehensive hardware and software setup is used, which not only includes tracking, but also an X3D based layer to simplify application development. Therefore it allows designers and other non-programmers to develop multi-touch applications very efficiently, while allowing to focus on user interaction and content.

IEEE Systems, Man and Cybernetics Society: IEEE International Conference on Systems, Man and Cybernetics : SMC 2009. New York: IEEE Press, 2009, pp. 565-570

This paper presents a simple yet effective approach for classification of human postures by using a time-of-flight camera. We investigate and adopt linear projection techniques such as Locality Preserving Projections (LPP), Linear Discriminant Analysis (LDA) and Principal Component Analysis (PCA), which are more widespread in face recognition and other pattern recognition tasks.We analyze the relations between LPP and LDA and show experimentally that using LPP in a supervised manner effectively yields very similar results as LDA, implying that LPP may be regarded as a generalization of LDA. Features for offline training and online classification are created by adopting common image processing techniques such as background-subtraction and blob detection to the time-of-flight data.

Sablatnig, Robert (Ed.) et al.: 15th International Conference on Virtual Systems and Multimedia. Proceedings : VSMM 2009. Los Alamitos, Calif.: IEEE Computer Society, 2009, pp. 193-196

In this paper we are presenting Cultural Heritage Layers, an approach that enables the visualization of historic media like drawings, paintings and photographs of buildings and historic scenes seamlessly superimposed on reality via video see through using X3D. This enables simple, inexpensive and sustainable Augmented Reality applications in the cultural heritage and architectural area based on industry standards. The main idea is to use existing historic media from archives and superimpose them seamlessly on reality at the right spot. These locative layers are context sensitively telling the location's history and create the impression of a virtual time journey. The registration of the virtual objects in the video images is provided by a robust 6DOF tracking framework based on two technologies that work in tandem: an initialization step based on Randomized Trees and a frame-to-frame tracking phase based on KLT. The entire application runs in real time on current Ultra Mobile PCs and MIDs.

Debattista, Kurt (Ed.) et al.: VAST 2009. VAST-STAR, Short and Project Papers Proceedings. Msida: University of Malta, 2009, pp. 112-116

Museums often lack the possibilitiy to present archaeological or cultural heritage sites in a realistic and interesting way. Thus are proposing a new way to show augmented reality applications of cultural heritage sites at remote places like museums. In the exhibition space large wall-filling photographs of the real site are superimposed with interactive contextual annotations like 3D reconstructions, images and movies. Therefore we are using two different hardware setups for visualization: Standard UMPCs and a custom made revolving display. The setup has been installed and tested at SIGGRAPH 2008, Allard Pierson Museum in Amsterdam and CeBIT 2009. Museum visitors could experience Forum Romanum and Saticum in an informative and intuitive way by pointing the video see through devices on different areas of the photographs. The result is a more realistic and entertaining way for presenting cultural heritage sites in museums. Furthermore our solution is less expensive than comparable installations regarding content and hardware.

Spencer, Stephen N. (Ed.): Proceedings WEB3D 2008 : 13th International Symposium on 3D Web Technology. New York: ACM Press, 2008, pp. 27-30

Multi-touch interaction on tabletop displays is a very active field of todays HCI research. However, most publications still focus on tracking techniques or develop a gesture configuration for a specific application setup. Very few explore generic high level interfaces for multi-touch applications. In this paper we present a comprehensive hardware and software setup, which includes an X3D based layer to simplify the application development process. We present a robust FTIR based optical tracking system, examine in how far current sensor and navigation abstractions in the X3D standard are useful and finally present extensions to the standard, which enable designers and other non-programmers to develop multi-touch applications very efficiently.

Diese Arbeit behandelt die Rekonstruktion von monokularen Bildfolgen mittels Kanten. Die Fehlertolleranz soll dabei durch die Verwendung von Snakes, zum Tracken der Kanten, erheblich verbessert werden. Geprüft wurde ob und in welchem Maß dieser Ansatz zu einer Verbesserung bzw. Optimierung führt. In dem dabei entstandenen Plugin für die Visionlib des Fraunhofer IGD werden die Snakes direkt auf den Kantenfeatures initialisiert. Mittels Gradientenwerten, Spannung und Krümmung bleiben die Snakes auf Features durch die Sequenz erhalten. Ausreißer werden erkannt und zurück auf das entsprechende Feature geführt, ohne das eine neue Snake erstellt werden muss. Die Ergebnisse zeigen, dass der Energieterm der Snakes allein schon zu einer signifikanten Verbesserung der Robustheit führt, da einmal begangene Fehler durch weitere Iterationen wieder behoben werden können. Die Präzision kann derweil mit anderen Verfahren mithalten und liegt bei einer Abweichung von ca. 1-2 Grad. Snakes bieten somit eine effiziente Möglichkeit das Tracking von Kanten zu optimieren, ihr Potential sollte in diesem Bereich weiter ausgeschöpft werden.

One of the key problems of augmented reality is the tracking of the camera position and viewing direction in real-time. Current vision-based systems mostly rely on the detection and tracking of fiducial markers. Some markerless approaches exist, which are based on 3D line models or calibrated reference images. These methods require a high manual preprocessing work step, which is not applicable for the efficient development and design of industrial AR applications. The problem of the preprocessing overload is addressed by the development of vision-based tracking algorithms, which require a minimal workload of the preparation of reference data. A novel method for the automatic view-dependent generation of line models in real-time is presented. The tracking system only needs a polygonal model of a reference object, which is often available from the industrial construction process. Analysis-by-synthesis techniques are used with the support of graphics hardware to create a connection between virtual model and real model. Point-based methods which rely on optical flow-based template tracking are developed for the camera pose estimation in partially known scenarios. With the support of robust reconstruction algorithms a real-time tracking system for augmented reality applications is developed, which is able to run with only very limited previous knowledge about the scene. The robustness and real-time capability is improved with a statistical approach for a feature management system which is based on machine learning techniques.

Bebis, George (Ed.) et al.: Advances in Visual Computing. Proceedings Part I : ISVC 2008. Berlin, Heidelberg, New York: Springer, 2008. (Lecture Notes in Computer Science (LNCS) 5358), pp. 530-539

Camera-based tracking systems which reconstruct a feature map with structure from motion or SLAM techniques highly depend on the ability to track a single feature in different scales, different lighting conditions and a wide range of viewing angles. The acquisition of high quality features is therefore indispensable for a continuous tracking of a feature with a maximum possible range of valid appearances. We present a tracking system where not only the position of a feature but also its surface normal is reconstructed and used for precise prediction and tracking recovery of lost features. The appearance of a reference patch is also estimated sequentially and refined during the tracking, which leads to a more stable feature tracking step. Such reconstructed reference templates can be used for tracking a camera pose with a great variety of viewing positions. This feature reconstruction process is combined with a feature management system, where a statistical analysis of the ability to track a feature is performed, and only the most stable features for a given camera viewing position are used for the 2D feature tracking step. This approach results in a map of high quality features, where the real time capabilities can be preserved by only tracking the most necessary 2D feature points.

Ashley, Michael (Ed.) et al.: VAST 2008. Proceedings. Aire-la-Ville: Eurographics Association, 2008, pp. 71-77

We present Reality Filtering, an application that makes it possible to visualize original content like drawings or paintings of buildings and frescos seamlessly superimposed on reality by using filtered augmented reality. This enables simple and inexpensive applications in the cultural heritage and architecture area. The main idea is that the video stream showing the reality is filtered on the fly to acquire the same presentation style as the virtual objects. This allows for a better integration of original historic content and creates the impression of a virtual time journey. The registration of the virtual objects in the video images is provided by a robust 6DOF tracking framework based on two technologies that work in tandem: an initialization step based on Randomized Trees and a frame-to-frame tracking phase based on KLT. For the initialization, we present the novel concept of temporally distributed computational load (TDCL), which is able to automatically detect and register multiple objects while maintaining a constant video frame rate of 20 frames / sec. For mid- to long-range augmentation a pure 2- dimensional tracking with 3DOF is applicable and leads to significant performance gain. The entire application runs in real time on Ultra Mobile PCs.

Institute of Electrical and Electronics Engineers (IEEE): Proceedings of 3DTV-CON 2007 [CD-ROM] : Capture, Transmission and Display of 3D Video, 4 p.

In this paper we introduce a novel architecture for rapid development and assessment of advanced 3D visual tracking systems. Indeed, we notice that it does not exist up to now a universal tracking approach that fulfils the requirements of all possible application scenarios at the same time. On contrary, very specific and performing solutions can be developed for given situations and uses. Therefore, software for visual tracking must be designed as a highly flexible system that can be quickly re-configured in order to enable the development of optimised solutions in terms of accuracy, robustness, frame rate and delay to this purpose we designed an architecture that offers many functionalities which can be combined together, and thus build a new processing chain. The overall system offers numerous advantages, such as interactive programming, real-time access to the data and parameter at runtime.

International Association for Pattern Recognition (IAPR): ICVS 2007 - Vision Systems in the Real World [Online] : Adaptation, Learning, Evaluation [online]. [cited 14 June 2007] Available from: http://www.icvs2007.org/programme.php 2007, 10 p.

In Augmented Reality applications, a 3D object is registered with a camera and visual augmentations of the object are rendered into the users field of view with a head mounted display. For correct rendering, the 3D pose of the users view w.r.t. the 3D object must be registered and tracked in real-time, which is a computational intensive task. This contribution describes a distributed system that allows to track the 3D camera pose and to render images on a light-weight mobile front end user interface system. The front end system is connected by WLAN to a backend server that takes over the computational burden for real-time tracking. We describe the system architecture and the tracking algorithms of our system.

IEEE Computer Society: ISMAR 2007 : Proceedings of the Sixth IEEE and ACM International Symposium on Mixed and Augmented Reality. Los Alamitos, Calif.: IEEE Computer Society, 2007, pp. 281-282

Since the last ten years product development in automotive industry is changing radically. Most physical mock-ups have vanished and are now replaced by digital ones. But they are still needed for final evaluations or issues, which cannot be adequately simulated. During their production, deviations from the CAD model may be made. Since digital and real mock-up must match for the further product development, the transfer of differences between physical and digital mock-up to the CAD format is a crucial issue. In this paper an Augmented Reality (AR) based tool-chain is presented, which allows matching the CAD data with real mock-ups and documents the differences between them. Essential functions like measurement and online construction are provided, allowing the end-users to create information in AR space and feeding them back into the CAD model.

Yagi, Yasushi (Ed.) et al.: Computer Vision - ACCV 2007. Berlin; Heidelberg; New York: Springer, 2007. (Lecture Notes in Computer Science (LNCS) 4843), LNCS 4843, pp. 769-778

In dynamic scenes with occluding objects many features need to be tracked for a robust real-time camera pose estimation. An open problem is that tracking too many features has a negative effect on the real-time capability of a tracking approach. This paper proposes a method for the feature management which performs a statistical analysis of the ability to track a feature and then uses only those features which are very likely to be tracked from a current camera position. Thereby a large set of features in different scales is created, where every feature holds a probability distribution of camera positions from which the feature can be tracked successfully. As only the feature points with the highest probability are used in the tracking step, the method can handle a large amount of features in different scale without losing the ability of real time performance. Both the statistical analysis and the reconstruction of the features' 3D coordinates are performed online during the tracking and no preprocessing step is needed.

Journal of Virtual Reality and Broadcasting, Vol.4 (2007), 1, 9 p.

In this paper we present a model-based approach for real-time camera pose estimation in industrial scenarios. The line model which is used for tracking is generated by rendering a polygonal model and extracting contours out of the rendered scene. By un-projecting a point on the contour with the depth value stored in the z-buffer, the 3D coordinates of the contour can be calculated. For establishing 2D/3D correspondences the 3D control points on the contour are projected into the image and a perpendicular search for gradient maxima for every point on the contour is performed. Multiple hypotheses of 2D image points corresponding to a 3D control point make the pose estimation robust against ambiguous edges in the image.

Kropatsch, Walter G. (Ed.) et al.: Computer Analysis of Images and Patterns. Proceedings : CAIP 2007. Berlin, Heidelberg, New York: Springer, 2007. (Lecture Notes in Computer Science (LNCS) 4673), pp. 20-27

In this paper we present a novel analysis-by-synthesis approach for real-time camera tracking in industrial scenarios. The camera pose estimation is based on the tracking of line features which are generated dynamically in every frame by rendering a polygonal model and extracting contours out of the rendered scene. Different methods of the line model generation are investigated. Depending on the scenario and the given 3D model either the image gradient of the frame buffer or discontinuities of the z-buffer and the normal map are used for the generation of a 2D edge map. The 3D control points on a contour are calculated by using the depth value stored in the z-buffer. By aligning the generated features with edges in the current image, the extrinsic parameters of the camera are estimated. The camera pose used for rendering is predicted by a line-based frame-to-frame tracking which takes advantage of the generated edge features. The method is validated and evaluated with the help of ground-truth data as well as real image sequences.

Institute of Electrical and Electronics Engineers (IEEE): ISMAR 2006 : Proceedings of the Fifth IEEE and ACM International Symposium on Mixed and Augmented Reality. Los Alamitos, Calif.: IEEE Computer Society, 2006, pp. 56-65

One of the key requirements of augmented reality systems is a robust real-time camera pose estimation. In this paper we present a robust approach, which does neither depend on of.ine pre-processing steps nor on pre-knowledge of the entire target scene. The connection between the real and the virtual world is made by a given CAD model of one object in the scene. However, the model is only needed for initialization. A line model is created out of the object rendered from a given camera pose and registrated onto the image gradient for finding the initial pose. In the tracking phase, the camera is not restricted to the modeled part of the scene anymore. The scene structure is recovered automatically during tracking. Point features are detected in the images and tracked from frame to frame using a brightness invariant template matching algorithm. Several template patches are extracted from different levels of an image pyramid and are used to make the 2D feature tracking capable for large changes in scale. Occlusion is detected already on the 2D feature tracking level. The features' 3D locations are roughly initialized by linear triangulation and then refined recursively over time using techniques of the Extended Kalman Filter framework. A quality manager handles the influence of a feature on the estimation of the camera pose. As structure and pose recovery are always performed under uncertainty, statistical methods for estimating and propagating uncertainty have been incorporated consequently into both processes. Finally, validation results on synthetic as well as on real video sequences are presented.

In dieser Diplomarbeit wird das Problem der Rekonstruktion von 3D-Linienmodellen aus Videodaten behandelt. 3D-Rekonstruktion ist ein interessantes Forschungsgebiet, das seit den 1970er Jahren große Beachtung findet. Ziel dieser Diplomarbeit war es dabei, ein System zur vollautomatischen Generierung von Linienmodellen zu erstellen, welche später zum markerlosen Tracking verwendet werden können. Die automatische Generierung hat dabei den Vorteil, daß dieselben Verfahren zur Kantenerkennung zum Einsatz kommen wie später beim Tracking. Weiterhin wurde ein Verfahren zum inkrementellen Update der bestehenden Rekonstruktion aus weiteren Videodaten entwickelt und evaluiert. Die Ergebnisse haben gezeigt, daß das Problem generell lösbar ist, aber auch, daß es noch viel Verbesserungsbedarf hinsichtlich Geschwindigkeit und Stabilität gibt. Es existieren verschiedene Ansätze zur 3D-Rekonstruktion, die bedeutendsten davon sind Faktorisierungsansätze -- meist gefolgt von Bündelblockausgleich (bundle adjustment) -- und Verfahren über die Fundamentalmatrix (two-view, Punktfeatures) beziehungsweise den Trifokaltensor (three-view, Linienfeatures). In der Diplomarbeit wird aus mehreren Gründen das Verfahren über den Trifokaltensor gewählt. Der Algorithmus wurde unter Zuhilfenahme der VisionLib-Bibliothek implementiert, die in der Abteilung A4 des Fraunhofer IGD entwickelt wird. Die Implementierung wurde mit Hilfe von synthetischen und realen Daten evaluiert. Die Ergebnisse waren vielversprechend, es gelangen gute Rekonstruktionen auf Basis von realen Videodaten. Allerdings zeigte sich auch, daß es noch Probleme hinsichtlich Stabilität und Geschwindigkeit gibt.

Hochschule für Technik Stuttgart: 1. Internationales Symposium "Geometrisches Modellieren, Visualisieren und Bildverarbeitung". Proceedings. Stuttgart, 2006, pp. 105-112

Dieses Paper beschreibt mehrere bildbasierte Trackingmethoden, welche für industrielle Augmented Reality (AR) Anwendungen genutzt werden können. Jeder der vorgestellten Algorithmen hat Stärken und Schwächen und daher ist keines dieser Verfahren für alle möglichen Szenarien geeignet. Vielmehr ergänzen sich die verschiedenen Verfahren, so dass von jedem Algorithmus die Stärken genutzt werden können und damit ein robustes und echtzeitfähiges Tracking möglich ist. Es werden folgende verschiedenen Stadien des Trackings beschrieben: Die Initialisierung zur Bestimmung der ersten Kameraposition, das Tracken von Bild zu Bild, und die Reinitialisierung nach dem gescheiterten Tracking.

Müller, Stefan (Ed.) et al.: Virtuelle und Erweiterte Realität : 3. Workshop der GI-Fachgruppe VR/AR. Aachen: Shaker, 2006. (Berichte aus der Informatik), pp. 155-164

In this paper we present a model-based approach for real-time camera pose estimation in industrial scenarios. The line model which is used for tracking is generated by rendering a polygonal model and extracting contours out of the rendered scene. By un-projecting a point on the contour with the depth value stored in the z-buffer, the 3D coordinates of the contour can be calculated. For establishing 2D/3D correspondences the 3D control points on the contour are projected into the image and a perpendicular search for gradient maxima for every point on the contour is performed. Multiple hypotheses of 2D image points corresponding to a 3D control point make the pose estimation robust against ambiguous edges in the image.

Kuhlen, Torsten (Ed.) et al.: Virtuelle und Erweiterte Realität : 2. Workshop der GI-Fachgruppe VR/AR. Aachen: Shaker, 2005. (Berichte aus der Informatik), pp. 107-118

In this paper, we present an overview of several visual tracking methods for industrial augmented reality applications. We show that no universal algorithm can deal with the large number of possible scenes, and that the different methods have to be seen as complementary approaches that all have their strengths and weaknesses. The main difficulty, then, consists in combining existing building blocks in the right manner so that the overall system enables stable tracking. This paper addresses each phase of the tracking, i.e. Initialization, Tracking, Re-Initialization, and proposes a first choice of appropriate algorithms. Finally, a global system is designed, tested and evaluated with help of video sequences of different real environments.

Dieses Paper beschreibt mehrere bildbasierte Trackingmethoden, welche für industrielle Augmented Reality (AR) Anwendungen genutzt werden können. Jeder der vorgestellten Algorithmen hat Stärken und Schwächen und daher ist keines dieser Verfahren für alle möglichen Szenarien geeignet. Vielmehr ergänzen sich die verschiedenen Verfahren, so dass von jedem Algorithmus die Stärken genutzt werden können und damit ein robustes und echtzeitfähiges Tracking möglich ist. Es werden folgende verschiedenen Stadien des Trackings beschrieben: Die Initialisierung zur Bestimmung der ersten Kameraposition, das Tracken von Bild zu Bild, und die Reinitialisierung nach dem gescheiterten Tracking.

Institute of Electrical and Electronics Engineers (IEEE): ISMAR 2005 : Proceedings of the Fourth IEEE and ACM International Symposium on Mixed and Augmented Reality. Los Alamitos, Calif.: IEEE Computer Society, 2005, pp. 62-69

We present a real-time model-based line tracking approach with adaptive learning of image edge features that can handle partial occlusion and illumination changes. A CAD (VRML) model of the object to track is needed. First, the visible edges of the model with respect to the camera pose estimate are sorted out by a visibility test performed on standard graphics hardware. For every sample point of every projected visible 3D model line, a search for gradient maxima in the image is then carried out in a direction perpendicular to that line. Multiple hypotheses of these maxima are considered as putative matches. The camera pose is updated by minimizing the distances between the projection of all sample points of the visible 3D model lines and the most likely matches found in the image. The state of every edge's visual properties is updated after each successful camera pose estimation.We evaluated the algorithm and showed the improvements compared to other tracking approaches