Instant Points Fast Rendering of Unprocessed Point Clouds.pdfVIP

Eurographics Symposium on Point-Based Graphics (2006) M. Botsch, B. Chen (Editors) Instant Points: Fast Rendering of Unprocessed Point Clouds Michael Wimmer? and Claus Scheiblauer Vienna University of Technology, Austria Abstract We present an algorithm to display enormous unprocessed point clouds at interactive rates without requiring long postprocessing. The novelty here is that we do not make any assumptions about sampling density or availability of normal vectors for the points. This is very important because such information is available only after lengthy postprocessing of scanned datasets, whereas users want to interact with the dataset immediately. Instant Points is an out-of-core algorithm that makes use of nested octrees and an enhanced version of sequential point trees. Categories and Subject Descriptors (according to ACM CCS): I.3.3 [Computer Graphics]: Display algorithms, View- ing algorithms 1. Introduction Point-based rendering has gained a lot of popularity due to the availability of 3D range scanners. Recent long-range laser scanners capture data up to 1000m, giving wildly vary- ing sampling densities (Fig. 1). Practically all current point rendering approaches rely on assumptions about sampling densities, or the availability of normal vectors for each point. In fact, most papers present models that have been gained by point sampling a geometric mesh. Unfortunately, post-processing a scanned point cloud to obtain a mesh can take several person months. This is because especially data acquired in the outdoors is not amenable to automatic postprocessing methods. Such data is characterized by holes in almost any surface, varying sam- pling densities, and a mixture of surface and non-surface structures (such as leaves), which is a problem for mesh- ing approaches. There are several efficient methods to esti- mate the normal vectors required for point rendering tech- niques directly from the point cloud [TKDS05]. However, normal estimation assumes that th