The Rendering Of Point Cloud Based On Ray Tracing And Photon Mapping

Rendering of realistic graphics is one of main research contents of comptergraphics, and it pursues realistic rendering of scenes. The usual ray tracing methods ofpoint models have some drawbacks such as coarse silhouettes and vague details.Photon mapping based on photon maps is a highly effective method producing globalillumination. But number of photons of photon map is limited because of limitation ofmemory space, which easily leads to bias and noise of illumination estimation.Progressive photon mapping can effectively reduce bias and noise in illuminationestimation by many times of iterative computations, but this needs very long photontracing stages.This dissertation does the further research on ray tracing of point models andphoton mapping based on photon maps, and its research results are listed below:1. A height-difference-based ray tracing of point models. To splat-based pointmodels, this paper theoretically analyzes the reason for which implicit surface methodhas bad rendering effects (i.e. coarse silhouettes and non-clear details etc.), wherejoint part between splats is non-coincident with geometric feature of splats. This paperproposes a height-difference-based ray tracing method that iteratively computesintersections by taking height difference between iterative point and local surface asstep length. And height difference is computed by blending projective distancesbetween the point of ray and the corresponding splats. Our method can ensure ourrendering effect close to true geometric feature of local surface, and has betterrendering effects than the ray tracing method based on implicit surface in thesilhouettes and details of point models.2. A K-neighbor-based ray tracing of point models. To rude point models withonly3D coordinates, we propose a K-neighbor-based ray tracing method, whichincludes basic method and improved method. We do an iterative intersectioncomputation between a ray and local plane corresponding to the actual K nearestpoints to iterative point of ray. When intersection conditions between ray and localplane is arrived, basic method takes iterative point and normal of local plane ascoordinates and normal of intersection, while improved method gets more precisecoordinates and normal of intersection by regular sampling on local surface.Compared with our basic method, our improved method can produce better smoothrendering effects for point models with some noise and more geometry details, andensure rendering effect of geometry details. Compared with the usual ray tracingmethods of splat-based point models, our improved method has better renderingeffect.3. A space-distribution-based reconstruction of global photon map and a geometry-bias-based illumination estimation of photon map. On the one hand, wepropose a space-distribution-based reconstruction of global photon map, this methodcan effectively reduce noise and geometric bias of illumination estimation, andenhance direct focusing illumination (caustics) and indirect focusing illuminations byglobal photon map after reconstruction of photon map. On the other, we propose ageometry-bias-based illumination estimation of photon map. Based on true area ofillumination estimation, which is intersection area between bounding sphere ofphotons estimating illumination and almost coplanar triangle meshes including thesephotons, the estimated illumination effect can be largely improved by reduction ofgeometry bias of illumination estimation.4. Photon mapping based on noise-reduced photon maps. We proposenoise-reduced photon map, which is different from photon maps produced by themethod of density control for photon map or the method of hierarchical photonmapping. And energies of all photons of noise-reduced photon map are largelyreduced. Under the condition of limitation of memory space, progressive photonmapping make illumination values of picture close to true illumination value by manyiterative steps, but they have to repeat very long photon tracing stage when changingviewpoint of scene. This paper proposes photon mapping based on noise-reducedphoton maps. In photon tracing stages, we produce some noise-reduced photon mapsby energy distribution between photons and save them into external memory. Afterthese noise-reduced photon maps are saved to external memory, we need not do verylong photon tracing stages again, and load these photon maps into memory which areused to produce the corresponding pictures by illuminatin estimation, the final pictureis the average of these pictures. This can effectively reduce bias and noise ofillumination estimation. Our method include basic methd and extended method,compared with our basic method, our extended method can more effectively reducethe bias of final image by increasing the photon map number of each group ofnoise-reduced photon maps.Our research has important application prospects, the corresponding technologiesand methods refer to computer graphics, rendering of realistic graphics and imageprocessing and et al. In aspects of ray tracing based on point models and globalillumination based on photon mapping, we propose the corresponding novel andpractical algorithms.