693 
REAL-TIME RENDERING OF LARGE TERRAIN ON MOBILE DEVICE 
JiangWen 8 , BaoshanZhu 8 , FanWang 8 
8 Zhengzhou Institute of Surveying and Mapping, No.66, Longhai Road, Zhengzhou, China - 
kissfro9642@sina.com 
Commission VI, WG VI/4 
KEY WORDS: Terrain rendering, Multi-resolution, Level of detail, Quad-tree, Adaptive rendering, Mobile device 
ABSTRACT: 
Terrain rendering is an important factor in representation of virtual scene. If terrains are large and detailed, there will be huge amount 
of data, so it is necessary to reduce the complexity of the rendered scene in real-time on mobile device. This paper proposes a 
multi-resolution technique to simplify the scenes and improve the speed of terrains rendering. Firstly, the full terrain height-field is 
divided into regular tiles, and then the appropriate level of detail is computed and generated dynamically, allowing for smooth 
changes of resolution across area of the surface. Each visible tile is then rendered using a computed triangle strip in an adaptive way 
according to viewpoint. The method is different from the triangle-based LOD algorithms and is optimized for modem to minimize 
CPU usage during rendering. The key of the technique is to develop an adaptive LOD framework that can optimally feed the graphic 
pipeline. At last, this paper also proposes a method of removing cracks on the meshes boundary. 
1. INTRODUCTION 
Terrain rendering on mobile devices plays an essential role in 
wide range of applications such as video games, virtual reality, 
3D environmental analysis, personal navigation and many 
geographic information system (GIS) applications. Despite 
mobile devices have seen dramatic improvements in last few 
years, the mobile devices are still clearly less capable than 
desktop computers in many ways. They run at a lower speed, 
display in smaller size and have lower resolution, there is less 
memory for running the programs and for storing them, and the 
battery of device can not last for long. 
On the other hand, terrain data obtained from the natural 
environment is usually very huge, and rendering accurate terrain 
implies the manipulation of very massive data sets which may 
contain billions of samples (e.g. triangles, points .etc.) and all 
those samples must be computed one by one instantly. In 
general, real-time rendering of three dimensional computer 
graphics requires faster than 15 frames pre second (FPS). Such 
a complexity introduces two main limitations: it might not be 
possible to store the entire data sets in random-access memory 
(RAM) and/or to perform its rendering in real-time on mobile 
device. 
In fact, rendering 3D terrains on mobile devices is still a very 
complex task because of the vast computational power required 
to achieve a usable performance. There are many visualization 
techniques that have been developed for PCs and workstations. 
However, using these same approaches for mobile devices 
introduces some unresolved problems. Most of the existing 
approaches are simplification of a triangulated model that 
represents terrain surface. Some solutions entirely rely on CPU 
whereas others use both CPU and GPU (sometimes using 
programs). GPU are able to render millions of triangles per 
second and even more when using triangles strips. However the 
GPU accelerated method still requires the full CPU power to 
compute vertex indices at every frame, even if the method is 
very efficient, it is not practicable when targeting mobile 
devices. Indeed, although some recent mobile devices dispose 
of GPU, they are not yet programmable. 
Aim at the target of real-time large terrain rendering on mobile 
platform; this paper propose a dynamic adaptive 
multi-resolution modelling to represent terrain based on 
quad-tree. The solution can be decomposed in two main parts. 
The first one is pre-processing. The main purpose is to construct 
multi-resolution digital elevation models (DEM) to represent 
terrains. The quad-tree structure is the key point in this part. The 
second one is dedicate to render a maximum number of 
triangles in view region. The purpose of this part is reducing the 
triangles to be drawn and maintaining the largest polygon area 
around the viewpoint. By loading tiles which are in view-field 
into client memory dynamically, it can free some memory and 
reduce the burden of CPU. 
2. PREVIOUS WORKS 
There are two familiar methods in the terrain rendering domain. 
The first one brings together methods that have been designed 
for terrain models which fit in memory (level of details 
technique). The second one gathers the algorithms designed for 
the rendering of large terrain data which can not be loaded into 
memory completely (out-of-core techniques). 
2.1 Previous terrain LOD techniques 
Terrain LOD algorithms use a hierarchy of mesh refinement 
operations to adapt the surface, and the methods are widely used 
in large terrain rendering now. LOD can decimate polygons thus 
reducing complexity of computation without affecting the 
quality of scenes. There are two schemes to choose proper LOD. 
One is based on the flatness of terrain surface. Large and coarse 
meshes are used at even regions, while tiny and refined meshes 
are represented at fluctuant regions. The other is based on the 
distance away from the eyes’ position. Regions which are nearer 
to the eyes’ position are rendered in refined meshes and the 
further regions are rendered in coarse meshes.