UMD Theses and Dissertations
Permanent URI for this collectionhttp://hdl.handle.net/1903/3
New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a given thesis/dissertation in DRUM.
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
3 results
Search Results
Item Measuring Deformations and Illumination Changes in Images with Applications to Face Recognition(2012) Jorstad, Anne; Jacobs, David; Applied Mathematics and Scientific Computation; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This thesis explores object deformation and lighting change in images, proposing methods that account for both variabilities within a single framework. We construct a deformation- and lighting-insensitive metric that assigns a cost to a pair of images based on their similarity. The primary applications discussed will be in the domain of face recognition, because faces provide a good and important example of highly structured yet deformable objects with readily available datasets. However, our methods can be applied to any domain with deformations and lighting change. In order to model variations in expression, establishing point correspondences between faces is essential, and a primary goal of this thesis is to determine dense correspondences between pairs of face images, assigning a cost to each point pairing based on a novel image metric. We show that an image manifold can be defined to model deformations and illumination changes. Images are considered as points on a high-dimensional manifold given local structure by our new metric, where costs are based on changes in shape and intensity. Curves on this manifold describe transformations such as deformations and lighting changes to connect nearby images, or larger identity changes connecting images far apart. This allows deformations to be introduced gradually over the course of several images, where correspondences are well-defined between every pair of adjacent images along a path. The similarity between two images on the manifold can be defined as the length of the geodesic that connects them. The new local metric is validated in an optical flow-like framework where it is used to determine a dense correspondence vector field between pairs of images. We then demonstrate how to find geodesics between pairs of images on a Riemannian image manifold. The new lighting-insensitive metric is described in the wavelet domain where it is able to handle moderate amounts of deformation, and allows us to derive an algorithm where the analytic geodesics between images can be computed extremely efficiently. To handle larger deformations in addition to changes in illumination, we consider an algorithmic framework where deformations are modeled with diffeomorphisms. We present preliminary implementations of the diffeomorphic framework, and suggest how this work can be extended for further applications.Item Geometric Algorithms for Objects in Motion(2010) Friedler, Sorelle Alaina; Mount, David; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)In this thesis, the theoretical analysis of real-world motivated problems regarding objects in motion is considered. Specifically, four major results are presented addressing the issues of robustness, data collection and compression, realistic theoretical analyses of this compression, and data retrieval. Robust statistics is the study of statistical estimators that are robust to data outliers. The combination of robust statistics and data structures for moving objects has not previously been studied. In studying this intersection, we consider a problem in the context of an established kinetic data structures framework (called KDS) that relies on advance point motion information and calculates properties continuously. Using the KDS model, we present an approximation algorithm for the kinetic robust k-center problem, a clustering problem that requires k clusters but allows some outlying points to remain unclustered. For many practical problems that inspired the exploration into robustness, the KDS model is inapplicable due to the point motion restrictions and the advance flight plans required. We present a new framework for kinetic data that allows calculations on moving points via sensor-recorded observations. This new framework is one of the first within the computational geometry community to allow analysis of moving points without a priori knowledge of point motion. Analysis within this framework is based on the entropy of the point set's motion, so efficiency bounds are a function of observed complexity instead of worst-case motion. Analysis is also considered under the more realistic assumptions of empirical entropy and assumptions of limited statistical independence. A compression algorithm within this framework is presented in order to address the storage issues created by the massive data sets sensors collect. Additionally, we show experimentally that this framework and accompanying compression scheme work well in practice. In order to allow for retrieval of the collected and compressed data, we present a spatio-temporal range searching structure that operates without the need to decompress the data. In sum, the collection scheme, compression algorithm, theoretical analyses, and retrieval data structures provide a practical, yet theoretically sound, framework within which observations and analyses can be made of objects in motion.Item Techniques For Video Surveillance: Automatic Video Editing And Target Tracking(2009) El-Alfy, Hazem Mohamed; Davis, Larry S; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Typical video surveillance control rooms include a collection of monitors connected to a large camera network, with many fewer operators than monitors. The cameras are usually cycled through the monitors, with provisions for manual over-ride to display a camera of interest. In addition, cameras are often provided with pan, tilt and zoom capabilities to capture objects of interest. In this dissertation, we develop novel ways to control the limited resources by focusing them into acquiring and visualizing the critical information contained in the surveyed scenes. First, we consider the problem of cropping surveillance videos. This process chooses a trajectory that a small sub-window can take through the video, selecting the most important parts of the video for display on a smaller monitor area. We model the information content of the video simply, by whether the image changes at each pixel. Then we show that we can find the globally optimal trajectory for a cropping window by using a shortest path algorithm. In practice, we can speed up this process without affecting the results, by stitching together trajectories computed over short intervals. This also reduces system latency. We then show that we can use a second shortest path formulation to find good cuts from one trajectory to another, improving coverage of interesting events in the video. We describe additional techniques to improve the quality and efficiency of the algorithm, and show results on surveillance videos. Second, we turn our attention to the problem of tracking multiple agents moving amongst obstacles, using multiple cameras. Given an environment with obstacles, and many people moving through it, we construct a separate narrow field of view video for as many people as possible, by stitching together video segments from multiple cameras over time. We employ a novel approach to assign cameras to people as a function of time, with camera switches when needed. The problem is modeled as a bipartite graph and the solution corresponds to a maximum matching. As people move, the solution is efficiently updated by computing an augmenting path rather than by solving for a new matching. This reduces computation time by an order of magnitude. In addition, solving for the shortest augmenting path minimizes the number of camera switches at each update. When not all people can be covered by the available cameras, we cluster as many people as possible into small groups, then assign cameras to groups using a minimum cost matching algorithm. We test our method using numerous runs from different simulators. Third, we relax the restriction of using fixed cameras in tracking agents. In particular, we study the problem of maintaining a good view of an agent moving amongst obstacles by a moving camera, possibly fixed to a pursuing robot. This is known as a two-player pursuit evasion game. Using a mesh discretization of the environment, we develop an algorithm that determines, given initial positions of both pursuer and evader, if the evader can take any moving strategy to go out of sight of the pursuer, and thus win the game. If it is decided that there is no winning strategy for the evader, we also compute a pursuer's trajectory that keeps the evader within sight, for every trajectory that the evader can take. We study the effect of varying the mesh size on both the efficiency and accuracy of our algorithm. Finally, we show some earlier work that has been done in the domain of anomaly detection. Based on modeling co-occurrence statistics of moving objects in time and space, experiments are described on synthetic data, in which time intervals and locations of unusual activity are identified.