Research On Rendering Techniques Of Synthetic Scenes Using Depth-Panoramic Video

With the rapid development of geometric modeling technologies, we can easily construct complex scenes. Those scenes may contain,thousands of Millions triangles with rich texture details, which beyond the capability of commodity graphics hardware. Till now, interactively generating photorealistic images from complex scenes is still a challenging task.Traditional geometry-based rendering systems, which employ model simplification and visibility culling techniques, can achieve real time display rate. However, they usually sacrifice the visual fidelity of the generated images. Alternatively, Irr age-Based Rendering (IBR) techniques generate photorealistic images from recorded images captured from real world or synthetic scenes. Since the rendering speed of IBR techniques is independent of the geometric complexity of scenes, it can achieve real time rendering of complex synthetic scenes on common PC.Most of the existing IBR techniques have the following drawbacks. Firstly, they overlook the inherent limitation of the input image resolution in the rendering process, and the quality of the rendered image heavily relies on the resolution of the irput images. Secondly, they are mainly designed for data captured from real world. The representations they adopt are complicated, uneasy to acquire and compress. Due to their complexity, the rendering algorithms are relatively complex, and cannot fully exploit the programmability of modern graphics processing units (GPUs). Since commercial rendering software can generate photorealistic images, in this dissertation, we study image-based rendering techniques for synthetic scenes, which attempt to overcome the drawbacks of the existing ones. The main co itributions of our work include,(1) We propose a novel rendering technique by blending multiple depth images for generating view-dependent lighting effects. This technique is based on our 3D warping equation, triangle-mesh reconstruction algorithm of depth image and view space z-bias algorithm. It makes a smart use of the floating-point blending function of the rewest graphics hardware to accumulate fragments' color values. Our rendering technique turns real-time rendering of destination image from hundreds of source depth images into reality.(2) We propose a scalable image-based rendering framework for synthetic scenes. Our framework adopts Depth-Panoramic Video (DPV) as the plenoptic primitive. The composite structures formed by multiple DPVs are employed to represent the whole scene. To deal with the varieties of the processing power of GPU and CPU of different hosts (PC, mobile device), we present three IBR techniques/systems, namely Depth-Panoramic Video Loop Rendering, Path-Panoramic Video Loop Rendering, and one Server Pushing system using Path-Panoramic Video Hopping technique, for interactive display of synthetic scenes.Depth-Panoramic Video Loop Rendering technique requires both CPU and GPU should have strong power. It allows the virtual viewpoint to move cortinuously inside of the video loop. As rendering, it first determines the regions from the input th it have contributions to the destination image, and then uses our multiple-depth-images blending :echnique to generate the novel image. Path-Panoramic Video Loop Rendering technique has nic; features of both Concentric Mosaics and Unstructured Lumigraph Rendering. It only requires CPU should be powerful, and can rendernovel images of viewpoints located inside of the video loop. The server-pushing system is based on client-server architecture, and designed for mobile devices which have poor CPU and GPU capabilities. In the system, server process is responsible for sending the compressed data required for rendering the destination image to client process through wireless network. After receiving the compressed data, client process first decompresses them, and then carries out local rendering using image warping. Several queues are used to cache the received and decompressed data for improving the system's performance.(3)We propos...