Rendering Very Large, Very Detailed Terrains
Thomas Lauritsen, Steen Lund · 2005
Usually rendering very large, very detailed terrains has high storage and processing requirements, because huge amounts of data are involved. In this document a new approach to solve the problem of rendering very larger, very detailed terrains is presented. Our proposal is to split the terrain representation into a low detail terrain, which does not required much storage and at runtime extend this low detail terrain with high amounts of details when needed. Our algorithm is based on two existing level-of-detail algorithms presented by Ulrich in [20] and De Boer in [3]. We are using the chunked quadtree structure presented by Ulrich, but the simplification process has been replaced by the scheme presented by De Boer. Together these two algorithms constitutes a simple, efficient level-of-detail algorithm and is suitable for runtime addition of details. The details can be calculated in various ways; we have chosen to use fractals for this purpose. Some techniques to speed up detail calculation is also shown. Details are added to the terrain at runtime by extending the quadtree with new leaf nodes. The value of the height samples in the new nodes are generated as a combination of the calculated details and a subdivision of the existing height samples. The subdivision scheme used is the one presented by Kobbelt in [11]. Results gathered from our implementation of our method shows that storage and memory requirements are low. They also shows that our level-of-detail algorithm is performing very well and renders high quality images. As such, the proposed solution works well; by adding details to an otherwise low detail terrain at runtime makes the terrain appear highly detailed, while storage requirements stays low. Given the high performance of the level-ofdetail algorithm, our method is capable of rendering very large, very detailed terrains. i Preface This document presents the results of the midway project conducted by the