TheQuantized kd-Tree: Efficient RayTracing ofCompressed PointClouds ErikHubo* TomMertenst
TomHaber Philippe Bekaert · 2006
Bothraytracing andpoint-based representations provide meansto efficiently display verycomplex 3Dmodels. Computational efficiency hasbeenthemainfocus ofprevious workonraytracing point-sampled surfaces. Forverycomplex models efficient storage intheformofcompression becomes necessary inorder toavoid costly diskaccess. However, asraytracing requires neighborhood queries, existing compression schemes cannot beapplied because oftheir sequential nature. Thispaper introduces anovel acceleration structure called theQuantized kd-tree, whichoffers bothefficient traversal andstorage. Thegist ofournewrepresentation lies inquantizing thekd-tree splitting plane coordinates. Weshowthat theQuantized kd-tree reduces thememoryfootprint upto18times, notcompromising performance. Moreover, thetechnique canalso beemployed toprovide LOD (Level-Of-Detail) toreduce aliasing problems, withlittle additional storage cost.