Augmented reality applications on hand-held devices suffer from the limited available processing power. While methods to detect the location of artificially textured markers within the scene are com- monly used, geometric properties of three-dimensional objects are rarely exploited for object tracking. In order to track such geometry efficiently on mobile devices, existing methods must be adapted. By focusing on key behaviors of edge-based models, we present a sparse depth buffer structure to provide an efficient rasterization method. This allows the tracking algorithm to run on a single CPU core of a current-generation hand-held device, while requiring only minimal support from the GPU.

Real-time 3D applications have evolved to the point where they become increasingly realistic. Several aspects have been subject to extensive research in order to deliver the maximum amount of realism possible within a limited computation budged. Examples are rendering or physics. One aspect, however, has not gotten the attention needed despite its omnipresence in any application, namely camera control and parametrization. In most interactive applications such as games, the camera placement and parametrization is either user controlled or pre-scripted by an artist. Only few attempts have been made to automate and simulate realistic camera behavior, as it is generally a difficult task. In this dissertation, we attempt to attack these shortcomings on different levels.
In the first part of this thesis we present a real-time camera control system that uses a global planning algorithm to compute large, occlusion free camera paths through complex environments. The algorithm incorporates the visibility of a focus point into the search strategy, so that a path is chosen along which the focus target will be in view.
In the second part, this thesis deals with camera parametrization for controlled stereoscopic rendering. We present an automatic controller for camera convergence and interaxial separation that specifically addresses challenges in interactive 3D applications like games. In such applications, unpredictable viewer or object motion often compromises stereopsis due to excessive binocular disparities. We derive constraints on the camera separation and convergence that enable our controller to automatically adapt to any given viewing situation and 3D scene, providing an exact mapping of the virtual content into a comfortable depth range around the display.
Finally, the third part of the thesis approaches advanced camera parametrization and the reproduction of realistic color balancing effects. The input to our algorithm is a sparse set of desired color correspondences between a source and a target image. The global color space transformation problem is then solved by computing a smooth vector field in CIE Lab color space that maps the gamut of the source to that of the target. Furthermore, we show how the basic per-image matching can be robustly extended to the temporal domain. This extension renders our method extremely useful for automatic, consistent embedding of synthetic graphics in video, as required by applications such as augmented reality.

This paper addresses the problem of globally balancing colors between images. The input to our algorithm is a sparse set of desired color correspondences between a source and a target image. The global color space transformation problem is then solved by computing a smooth vector field in CIE Lab color space that maps the gamut of the source to that of the target. We employ normalized radial basis functions for which we compute optimized shape parameters based on the input images, allowing for more faithful and flexible color matching compared to existing RBF-, regression- or histogrambased techniques. Furthermore, we show how the basic perimage matching can be efficiently and robustly extended to the temporal domain using RANSAC-based correspondence classification. Besides interactive color balancing for images, these properties render our method extremely useful for automatic, consistent embedding of synthetic graphics in video, as required by applications such as augmented reality.

This paper presents a controller for camera convergence and interaxial separation that specifically addresses challenges in interactive stereoscopic applications like games. In such applications, unpredictable viewer- or object-motion often compromises stereopsis due to excessive binocular disparities. We derive constraints on the camera separation and convergence that enable our controller to automatically adapt to any given viewing situation and 3D scene, providing an exact mapping of the virtual content into a comfortable depth range around the display. Moreover, we introduce an interpolation function that linearizes the transformation of stereoscopic depth over time, minimizing nonlinear visual distortions. We describe how to implement the complete control mechanism on the GPU to achieve running times below 0.2ms for full HD. This provides a practical solution even for demanding real-time applications. Results of a user study show a significant increase of stereoscopic comfort, without compromising perceived realism. Our controller enables ‘fail-safe’ stereopsis, provides intuitive control to accommodate to personal preferences, and allows to properly display stereoscopic content on differently sized output devices.

We present a real-time camera control system that uses a global planning algorithm to compute large, occlusion free camera paths through complex environments. The algorithm incorporates the visibility of a focus point into the search strategy, so that a path is chosen along which the focus target will be in view. The efficiency of our algorithm comes from a visibility-aware roadmap data structure that permits the precomputation of a coarse representation of all collision-free paths through an environment, together with an estimate of the pair-wise visibility between all portions of the scene. Our runtime system executes a path planning algorithm using the precomputed roadmap values to find a coarse path, and then refines the path using a sequence of occlusion maps computed on-the-fly. An iterative smoothing algorithm, together with a physically-based camera model, ensures that the path followed by the camera is smooth in both space and time. Our global planning strategy on the visibility-aware roadmap enables large-scale camera transitions as well as a local third-person camera module that follows a player and avoids obstructed viewpoints. The data structure itself adapts at run-time to dynamic occluders that move in an environment. We demonstrate these capabilities in several realistic game environments.

This thesis presents approaches and actions taken to enable capturing of two-dimensional water phenomena. In order to cope with the three-dimensional nature of water the construction of a specifically designed tank is shown. Additionally to this tank, a modifiable wedge-shaped device was built to create interesting splashes upon impact with the liquid. Reducing the threedimensionality is not only based on the apparatus, but also on subsequent algorithmic steps. After eliminating illumination and recording difficulties that come along with a high-speed camera, the k-means segmentation algorithm, known from computer vision, is utilized to finally reduce the three-dimensional water appearance to two dimensions. In a final step, a graph data structure is introduced which is paired with a user-controllable tool in order to allow higher semantics to be added to the data.

This semester thesis presents approaches to extend Oskam’s global camera control in [Osk08] by dynamic objects. By inspecting the nodes of a precomputed roadmap datastructure and its connections for intersections with bounding spheres, we are able to make the global camera control algorithm become aware of moving obstacles in a 3D environment. Furthermore, to guarantee the continuity of visibility-ensuring properties of the preliminary work, we introduce a visibility cone structure to satisfy these requirements. Finally, adapting the iterative path postprocessing algorithm is necessary to make the extension fully comply with the user’s visual needs.

This thesis presents an algorithm for visibility transition planning that can compute large, occlusion free camera paths in real time through complex environments. The algorithm incorporates the visibility of a focus point into the optimally criteria, so that the chosen path strives to keep the focus target in view. The efficiency of the algorithm comes from a visibility roadmap data structure that allows the precomputation of a coarse representation of all collision-free paths through an environment, together with an estimate of the pair-wise visibility between all portions of the environment. The runtime system executes a path planning algorithm using the precomputed roadmap values to find a coarse path that is optimal in terms of visibility up to the resolution of the roadmap. Next, a more exact visibility estimate is determined by computing a sequence of occlusion maps along the coarse path. The same path-planning algorithm is executed on these occlusion maps to ensure optimal visibility on a fine scale. An iterative smoothing algorithm, together with a physically-based camera model, ensures that the path followed by the camera is smooth in both space and time. Finally, a dynamic camera controller is shown that encorporates the visibility transition planning into a real-time re-routing system that is able to follow a fast paced game character in a complex environment.

There exist many approaches to object recognition of image data. Most of the methods use a topdown approach to classify the content, as low-level information is often considered inapplicable or insufficient for this task. The method presented in this work, however, shows a way to exploit low-level image segmentation for the purpose of categorizing different object classes of still images. The key idea is to not only use one single segmentation, but a whole set of different parameterized segmentations of an image as basis for feature extraction. Segment boundaries are used as paths along which strings of feature vectors are drawn. These strings are pairwise aligned to create a scoring matrix, which can be transformed into a Mercer kernel for a standard SVM based classifier. Despite the inherit problems of low-level segmentations, this method performs very well on standard benchmark image databases and shows that segmentation indeed can be used for object recognition.
The first goal of this semester thesis was to create a Matlab framework that implements the proposed method for object recognition. With this framework as basis, several approaches to improve scoring results of pairwise alignments, and thus improve the categorization rate, need to be tested. Also the parameters used for the string alignment can have an impact on the performance as well. Therefore, the second goal was to examine different combinations of methods for feature extraction and string alignment with different parameter settings in order to find out which configurations lead to the highest retrieval rates.

The variety of computer game genres has grown very fast over the last decade. Environments have become bigger and more complex. Also the intricacy of game play has increased along with the geometry in a game scene. One problem, which comes along with this, is the control of the virtual camera. Different situations arise where a simple approach may fail when several properties of the camera, like low-frequency movement or penetration avoidance, need to be achieved. Also different types of games require different kinds of camera movements and controls to enhance the game play. Three types of game cameras are wide spread and cover most of the genres: First person camera, third person camera, and isometric camera.

This work presents a method to encapsulate a camera module that is capable of performing the three main camera types simultaneously and still maintains a clean and easy to use interface. A controller is shown which can be integrated into the game loop, consisting of different camera control routines by which a physical camera model is guided through the environment. A variant of the camera simulation loop is also introduced, which provides an interface for ray casting and bounding volume tests without the camera knowing about the surrounding geometry. This extended controller is exploited to make the camera automatically avoid line of sight blocker and collisions depending on the active camera control routine.