| dc.description.abstract | In augmented reality applications, the position and orientation of the observermust be estimated in order to create a virtual camera that renders virtual objectsaligned with the real scene. There are a wide variety of motion sensors availablein the market, however, these sensors are usually expensive and impractical. Incontrast, computer vision techniques can be used to estimate the camera poseusing only the images provided by a single camera if the 3D structure of thecaptured scene is known beforehand. When it is unknown, some solutions useexternal markers, however, they require to modify the scene, which is not alwayspossible. </p><p>Simultaneous Localization and Mapping (SLAM) techniques can deal withcompletely unknown scenes, simultaneously estimating the camera pose and the3D structure. Traditionally, this problem is solved using nonlinear minimizationtechniques that are very accurate but hardly used in real time. In this way, thisthesis presents a highly parallelizable random sampling approach based on MonteCarlo simulations that fits very well on the graphics hardware. As demonstratedin the text, the proposed algorithm achieves the same precision as nonlinearoptimization, getting real time performance running on commodity graphicshardware. </p><p>Along this document, the details of the proposed SLAM algorithm areanalyzed as well as its implementation in a GPU. Moreover, an overview of theexisting techniques is done, comparing the proposed method with the traditionalapproach. | en_US |
