中文摘要：

高度复杂的三维场景通常包含几千万甚至上亿个三角形和丰富的纹理,大大超过了目前图形硬件的处理能力。传统基于几何的绘制系统通过牺牲画面质量来实现场景的快速绘制。与此不同,基于图像的绘制技术利用逼真的图像序列来生成高质量的目标画面。研究面向虚拟场景的IBR（Image-Based Rendering）技术来克服现有方法的不足,提出以深度全景视频DPV作为场景表示的基本单元,通过多段深度全景视频组成的深度全景视频网络来表示虚拟场景的漫游区域,它允许视点在漫游平面的封闭区域内连续运动。绘制算法根据目标视点参数计算深度全景视频环中对目标图像有贡献的候选区域,综合利用GPU的强大处理能力和浮点格式的绘制目标,以及多幅深度图像混合绘制技术对候选区域进行绘制来生成目标图像。实验结果表明,深度全景视频绘制技术可实现大规模虚拟场景高质量实时绘制。

英文摘要：

High complex 3-D scenes usually contain thousands of millions of triangles and rich texture details,which are far beyond the capability of commodity graphics hardware.Traditional geometry-based rendering systems can achieve fast display of real time scene through sacrificing the visual fidelity of the generated images.Alternatively,the image-based rendering（IBR） technique takes photorealistic images＇ sequence as its input to generate high quality destination images.In this paper,a synthetic scenes-oriented IBR rendering technique is studied for overcoming the drawbacks of existing techniques.The depth panoramic video（DPV） is proposed to be the elementary primitive of scene representation.The DPV networks composed of multiple DPVs are employed to represent the walkthrough zone in whole synthetic scene.This representation allows the viewpoint to move continuously in closed regions of the walkthrough plane.When rendering,the proposed technique first determines the candidate regions from the DPV ring that have contributions to the destination image according to destination viewpoint parameters and then generates novel image by rendering the candidate regions.The algorithm comprehensively exploits the power of modern GPU processing capability that supports render target of floating-point texture,as well as our multiple-depth-images blended rendering technique.Experimental results illustrate that the DPV rendering technique is able to generate high quality images with real time performance for large-scale synthetic scenes.