Research On Dynamic Light Field Capture And3D Modeling

Gets a visual scene is the main approach for human perception of the objectiveworld, since the plenoptic function proposed by Adelson et al in1991formMassachusetts Institute of Technology, on high dimensional visual information in theacquisition and capture the plenoptic function has become a scene acquisition and hotresearch to capture the direction, and the birth of computer vision, computer graphics,signal processing, optical and other disciplines the depth of the new cross field--computational photography. Light field acquisition is the core and foundation ofscience of computational photography, it is based on conventional imagingmechanism, the plenoptic function of each dimension visual information acquisitionand capture, hardware design and software calculation combined, break through thetraditional photography collection limitation, high dimensional visual information inall directions, truly get scene.This paper further focuses on the key technologies and3D modeling sampled lightfield dynamic, the realization of a diameter of10meters,7meters high dome shapedmulti view seven dimensional dynamic light field acquisition system. This system canrealize the synchronous control of light and high speed camera and precise, dynamicmulti view light field light field acquisition, satisfy the multi object dynamic sceneacquisition, and the multi object scene recovery of dense3D modeling, multi view.Chapter1, introduction, introduces the domestic and foreign research institutionslight field acquisition and3D modeling technology research.Chapter2, introduces dynamic light field acquisition system designed in this paper,and the realization process of acquisition system architecture design. Steel framestructure of the system can be expanded and flexibility, can according to theacquisition of target surface complex degree of different random adjustment angle andthe light field; through a multi camera array shooting and light emitting diode fastchanging illumination, using distributed synchronous control, precise synchronizationchanges of light and multi view acquisition; according to the calibration problem ofmulti camera array collection system, software realization, the system has a solutionsatisfactory results.Chapter3, introduces the3D reconstruction algorithm based on contour. To achieve a fast for video image sequence of foreground background segmentation algorithm toget the contour map, according to the contour information, exact polyhedral visualconvex hull algorithm existing mature generating visual convex hull based on3Dmesh generation, and for visual convex hull optimization measures based on Poissondisk sampling. Due to the optimization of extraction of contour information accuratelyand point cloud generated3D model,3D reconstruction algorithm based on thecontour, the integrity and accuracy.Chapter4, introduces the implementation of multi view3D reconstructionalgorithm based on the point cloud. Three steps to achieve the algorithm mainlyincludes: the cloud point extraction, fusion, point cloud triangulation, main researchon occlusion, noise and non textured regions have the robustness of the matchingalgorithm, edge/hidden point detection and fusion, from the perspective of point clouddenoising, based on confidence information Poisson surface reconstruction,reconstruction of surface structured problems, reconstruction framework of pointcloud obtained results of three-dimensional reconstruction of high robust, qualitybased on satisfaction.Finally, Chapter5summarized the conclusions of the research and the prospect offuture research.