Professional Documents
Culture Documents
net/publication/220649970
CITATIONS READS
37 598
3 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Bisheng Yang on 13 March 2017.
Research Article
1. Introduction
Recent advances in three-dimensimal (3D) spatial data capture, such as high
resolution satellite images and laser scanning, as well as corresponding data
processing and modelling technologies have led to the generation of large amounts
of datasets on terrains, buildings, roads and other features. On the one hand, the
huge amountal data provides the basis for presenting the real world at a very
detailed level. On the other hand, efficient methods for visualizing large data
volumes need to be developed. In fact, the volume of captured terrain data is beyond
the real time rendering capability of a current workstation. For example, a terrain
can be represented by several million triangles. To narrow the gap between the huge
data volume from terrain models visualized and the rendering capabilities of current
workstations, the terrain models need to be dynamically represented at different
resolutions. As a result, they can be rapidly visualized at various resolutions on a
currently available workstation.
Since Clark (1976) described the benefits by representing objects within a scene at
several different resolutions, research has been focusing on developing methods for
constructing multi-resolution models in areas such as 3D data compression,
visualization of terrain, simplification of models and progressive transmission.
These have led to various recently developed algorithms and models for multi-
resolution models construction (Mahdi and Chris 1998, Eck et al. 1995, Shi et al. 2003,
Yang et al. 2005a, 2005b), terrain simplification (Gerstner 2003, Bjorke and Nilsen
2003) and progressive transmission over the World Wide Web (Park et al. 2001).
The existing algorithms and models for constructng multi-resolution models can
be divided into two categories: Grid-based multi-resolution algorithms and
Triangulated Irregular Network (TIN)-based multi-resolution algorithms. Figure 1
gives the examples of a TIN model and a Grid model. Grid-based multi-resolution
algorithms include the quad-tree algorithm (Roettger et al. 1998, Lee and Samet
2000) and the triangle bisect algorithm. Many methods have been developed based
on the triangle bisect algorithm, such as the adaptive quad-tree (Lindstrom et al.
1996), the ROAMing algorithm (Duchaineau et al. 1997), Right TIN model (Evans
et al. 1997), the longest edge bisection algorithm (Lindstrom and Pascucci 2002) and
the dynamic adaptive meshes (Cignoni et al. 2003). Moreover, the quad-tree
algorithm constructs multi-resolution models is easy to implement based on a
bottom-up approach, and the triangle bisect algorithm constructs multi-resolution
models based on a top-down approach. The principles of the quad-tree algorithm
and the triangle bisect algorithm are illustrated in figure 2.
There has been extensive research on TIN-based multi-resolution algorithms. A
review and analysis of the existing solutions have been addressed in Heckbert and
Garland (1997). At present, there are three categories of algorithms that pertain to
the multi-resolution TIN models. These are algorithms that (a) simplify a mesh by
removing vertices; (b) simplify a mesh by removing edges; and (c) simplify a mesh by
removing triangles (Ribelles et al. 2002). Other related research includes methods
based on the transformation of wavelets.
Figure 2. Principles of (a) the quad-tree algorithm; and (b) the triangle bisection algorithm.
areas are represented only by Grid-based models. The following rules are proposed
for the formal definitions of terrain models with TIN-based and Grid-based models.
N Rule 1: The number of rows and columns of the base Grid (M0) are 2n + 1.
N Rule 2: The number of rows and columns of tiles (mi) are 2k + 1 (k,n) if the
tiles are in the Grid representation, and the union of all tiles is the
S
n
whole terrain model, mi ~M0 .
N i~0
Rule 3: The representation of each tile is either a Grid or a TIN,
mi5(Grid|TIN).
N Rule 4: The intersection of two tiles minmj is either their boundary or Ø.
N Rule 5: If the triangle (Ai) exists, there is one and only one tile that contains it.
That is to say, if Ai [ mk and Ai [ mj , then k5j.
Rules 1 to 5 give strictly mathematical definitions for the terrain models
represented by TINs and Grids. According to rules 1 to 5, it is clear that a terrain is
divided into many tiles and the spatial relationships among the tiles conform to the
above rules. Hence, a terrain (M) can be represented as:
[
n
M~ mi , mi ~ðV , F ÞjðLx, Ly, Z, rowsize, columnsize, row, col Þ
i~0
Where mi, a sub-region of the whole region, is either a Grid-based tile or a TIN-
based tile, vi5(xi, yi, zi) and fi is the triangle of <vi, vj, vk>, rowsize and columnsize
are the resolution of grid (the width and height of minimum unit grid tile), and row
and col are the numbers of row and column grid tiles. zi is the elevation of vertex in
Grid tiles, and Lx and Ly are the coordinates of the left-down corner of Grid tiles.
/* Integration of Grids and TINs, basedgrid is the base Grid, features are boundaries of
man-made objects */
void IntegrationGridandTin(Grid basedgrid, Features features){
for(int i50; i,features.size(); i + + ){
if (feature[i].type55point)
InsertpointinGrid(feature[i]);/* insert a point into a grid and triangulate the grid */
if(feature[i].type55line)
InsertedgeinGrid(feature[i]);/* insert an edge into a grid and triangulate the grid */
}
}
1024 B. Yang et al.
The above pseudocode algorithm illustrates the key steps for integrating Grid-
based and TIN-based models. Once all the features are integrated into the base Grid
(M0), a terrain model with representation of TIN and Grid will be generated.
Figure 3 shows an example of a terrain represented by TINs and Grids.
The lowest resolution Grid tile consists of four points (two triangles). A finer
resolution Grid tile can be achieved by splitting the coarser triangle. Therefore, the
longest edge bisection algorithm is a top-down approach. Moreover, a binary tree is
generated during the procedure (figure 4(b)).
Figure 4. The longest edge bisection algorithm: (a) the concept of the algorithm; and (b) the
binary tree.
Figure 5. The iterative edge contraction algorithm: (a) the concept of the algorithm; and (b)
the binary tree.
1026 B. Yang et al.
The edge with the minimum distance to the average plane will be firstly popped from
the stack for potential collapse.
As the integrated terrain model consists of many tiles, the multi-resolution model
of each tile is constructed based on either the longest edge bisection algorithm or the
iterative edge contraction algorithm. According to the workflow of constructing the
multi-resolution terrain models, an overall multi-resolution terrain model can be
represented as
[n
M~ m0i
i~0
Then there is no ‘gap’ between the common boundaries. Otherwise, a gap will have
been created, leading to visual artifacts.
In this study, a viable solution is proposed to seal the ‘gap’ between the common
boundaries of adjacent tiles.
N Find the parent vertices of the inactive vertices;
N Split the parent vertices to make the child vertices active.
Figure 6. Active vertices and inactive vertices in the common boundaries of adjacent tiles.
Multi-resolution terrain models 1027
Take figure 6 as example, let the union of A1active and A2active be Aactive,
Aactive5A1active<A2active5{a1, a2, a3, a4, a5}. Thus, the inactive vertices in the
common boundaries of tile A1 and tile A2 are Aactive2A1active5{a4} and
Aactive2A2active5{a2}, respectively. According to the above solution, the parent
vertices of a4 and a1 will be split to active them.
Once an inactive vertex is active, it will be a vertex of one triangle in a multi-
resolution model. After all the inactive vertices are active, the set of active vertices in
the common boundaries of adjacent tiles are identical. Thus, the ‘gap’ will be sealed,
and a multi-resolution terrain model can be achieved. Figure 7 illustrates an example
after sealing the gap between the common boundaries of adjacent tiles.
In light of the methodology for constructing multi-resolution Grid and TIN
models, both the individual multi-resolution Grid models and TIN models generate
their own binary trees either when a triangle is split or an edge is collapsed during
their generation. These binary trees should be stored and managed for generating
the integrated models at various resolutions. To fulfil the requirement, a vertex tree
is proposed to manage all the binary trees generated from each TIN and Grid-based
tiles. Figure 8 illustrates an example of the vertex tree. Here, a virtual vertex is
inserted as the root node of the vertex tree. The vertices at the first level of the vertex
tree are marked with the vertex id (e.g. V1, V2,….Vn) and the vertex category (Grid
or TIN), which indicates the type of tile that the vertex belongs to. The vertices at
the other levels of the vertex tree are only marked with the vertex id.
Each binary tree in the vertex tree represents an operation of vertex split/edge
collapse. These binary trees record the topology among vertices, edges and
triangles, called vertex topological relationships. Vertex topological relationships
have an essential impact on data storage and speed efficiency in constructing a
multi-resolution terrain model. Xia et al. (1997) stored vertex topological relation-
ships explicitly, and Ei-Sana and Varshney (1999) improved storage by storing them
implicitly. In fact, Ei-Sana and Varshney’s approach is unnecessarily restrictive for
half-edge collapsing (De Floriani and Magillo 2001), and can prevent further
simplification of the original model.
To store vertex topological relationships with less memory space, uniform data
structure is proposed in this study according to the general bracket method
(Donaghey 1980). The data structure is detailed in the following section.
general bracket method, the vertex tree can be encoded by a set of recorded vertices
and a set of brackets (bitstream of 0 and 1). The bracket set records the implicit
connectivity and levels among the recorded vertices in the vertex tree. The
connectivity and levels among the vertices in the vertex tree can be dynamically
encoded by traversing the bracket sets with a depth first algorithm. Figure 10
demonstrates the encoding result at a specific running time.
Various resolution models can be dynamically generated by splitting/collapsing
different numbers of vertices in the vertex tree. However, randomly splitting the
vertex or collapsing the edge may cause an invalid topology (triangle foldovers) in
the multi-resolution model, especially in the TIN-based tiles. Therefore, the edge
collapses or the vertex splits during a running time must conform to certain rules.
Xia et al. (1997) proposed the following validity rules on vertex split and edge
collapse: vertex c can be collapsed to vertex p, only when the vertices p0,..,pn are
presented as the neighbours of p and c for display (see figure 11(a)). However, the
rigorousness of the rules needs to be further investigated. Although the vertex
neighbourhood relationship conforms to the above rule, triangle foldovers are still
inevitable when vertex c is collapsed to vertex p (as illustrated in figure 11(b)).
Yang et al. (2005b) proposed a set of constraint rules to overcome the deficiency
of the existing rules and avoid the problem of triangle foldovers. The new constraint
rules are as follows:
Constraint -1: Only when the influent region of vertex vb is a convex polygon, can vertex vb be
collapsed to vertex va. (The concept of influent region is illustrated in figure 12(a)).
Constraint-2: The vertex vb can be split safely only if the vertices of the influent region are in a
lower position level in the vertex tree compared with that of vertex vb.
Figure 10. Encoding the vertex tree according to the general bracket method.
1030 B. Yang et al.
Figure 11. Invalid operation of the edge collapse: (a) An axample of edge collapse; and (b)
invalid edge collapse
Figure 12. Triangle influence region and judgment of features: (a) The influent region of a
vertex; and (b) judging a feature edge.
The proofs of above constraint rules are detailed in Yang et al. (2005b). To
enhance feature preservation abilities of the multi-resolution models, an additional
rule is proposed as follows:
Constraint -3: An edge shared by two triangles is defined as a feature edge if the angle between
the two triangles is over a specified value, for example 45u. If one edge is a feature edge, it does
not allow contracting. Figure 12(b) illustrates an example of judging a feature edge.
When one operation is invoked to collapse on edge or to split one vertex, the
operation must conform to the above rules to overcome surface foldovers and retain
features in multi-resolution models. Therefore, the multi-resolution models with
various errors can be generated by collapsing different numbers of edges and
splitting triangles. The error has an important effect on the accuracy and
visualization effect of the multi-resolution terrain models.
models is related to the resolution of models. Screen space error metric (Lindstrom
et al. 1995) proves to be an effective way to control the error of multi-resolution
models. According to the screen space error metric, the error of a multi-resolution
model is brought about by the change in z direction (height error) of different
resolution models. When a triangle is split or an edge is collapsed, a height error is
generated. Figure 13 illustrates the height error in the Grid tiles and TIN tiles. For
Grid tiles (figure 13(b)), the height error h is
h~hi, j {ðhm,n zhs,t Þ 2, i~ðmzsÞ=2, j~ðnztÞ=2
When one edge is collapsed to a vertex in TIN tiles (figure 13(a)), the difference
between the distances of the edge and vertex to the average plane is calculated as the
height error.
As constructing multi-resolution models of each tile, the vertex with the minimum
height error in the Grid tiles and TIN tiles will be split and collapsed, respectively.
Thus, terrain models at various resolutions can be achieved by specifying different
height error thresholds. On the other hand, view-dependent multi-resolution models
can be generated by specifying lower height error thresholds for vertices near to the
viewer and higher error thresholds for vertices far to the viewer.
5.1 Comparison between the new method and the quadric error metrics based
iterative edge contraction algorithm
To test speed performance in constructing the multi-resolution terrain models,
various resolutions models of the study area were generated with the quadric error
metrics based iterative edge contraction algorithm (Garland 1999) and the new
Figure 13. Height error in the integrated multi-resolution model: (a) height error in a TIN
tile; and (b) height error in a Grid tile.
1032 B. Yang et al.
Table 1. A comparison between the new method and the quadric error metrics based iterative
edge contraction algorithm.
1.107 m and 1.036 m, respectively. Therefore, the new method can achieve a higher
accuracy for a multi-resolution terrain model compared with the quadric error
metrics based iterative edge contraction algorithm.
6. Conclusions
As a further development to the existing single model-based multi-resolution
representations: either Grid-based or TIN-based models, this paper has presented a
new method for constructing multi-resolution terrain models with both TIN and
Grid representations.
To construct multi-resolution terrain models combining TIN-based and Grid-
based multi-resolution algorithms, we have presented a set of rules for the formal
definitions of terrains with TINs and Grids and have developed a new and efficient
method for encoding and storing vertex topological relationships in multi-resolution
terrain models. Moreover, a set of new rules is defined to overcome invalid topology
1034 B. Yang et al.
Figure 15. The multi-resolution model generated by the new method with constraint rules
(height error51.2 m, triangle55000): (a) 3D visualization; and (b) 2D visualization (the red
lines denote the boundary of the road).
Multi-resolution terrain models 1035
Figure 16. The multi-resolution model generated by the new method without constraint
rules (height error51.2 m, triangle54809): (a) 3D visualization; and (b) 2D visualization (the
red lines denote the boundary of the road).
1036 B. Yang et al.
Acknowledgement
The work described in this paper was substantially supported by grants from the
National Natural Science Foundation of China (No. 40401051), the Research
Grants Council of the Hong Kong Special Administrative Region (Project
No. PolyU 3-ZB40) and The Hong Kong Polytechnic University (1.34.9709). The
constructive comments of three anonymous reviewers are greatly appreciated.
Multi-resolution terrain models 1037
References
BJØRKE, J.T. and NILSEN, S., 2003, Wavelets applied to simplification of digital terrain
models. International Journal of Geographical Information Science, 17, pp. 601–621.
BONNEAU, G.P., 1998, Multiresolution analysis on irregular surface meshes. IEEE
Transactions on Visualization and Computer Graphics, 4, pp. 365–378.
CIAMPALINI, A., CIGNONI, P., MONTANI, C. and SCOPIGNO, R., 1997, Multiresolution
decimation based on global error. The Visual Computer, 13, pp. 228–246.
CIGNONI, P., GANOVELLI, F., GOBBETTI, E., MARTON, F., PONCHIO, F. and SCOPIGNO, R.,
2003, BDAM: batched dynamic adaptive meshes for high performance terrain
visualization. Computer Graphics Forum, 22, pp. 505–514.
CLARK, J., 1976, Hierarchical geometric models for visible surface algorithms. Communication
of ACM, 19, pp. 547–554.
DE FLORIANI, L. and MAGILLO, P., 2001, Multiresolution meshes, principles of multi-
resolution in geometric modeling. PRIMUS01 Summer School, 22–30 August 2001,
Munich, Germany, pp. 193–234.
DE FLORIANI, L., MARZANO, P. and PUPPO, E., 1996, Multiresolution models for topographic
surface description. The Visual Computer, 12, pp. 317–345.
DE FLORIANI, L., MAGILLO, P. and PUPPO, E., 2000, VARIANT: a system for terrain
modeling at variable resolution. GeoInformatica, 4, pp. 287–315.
DONAGHEY, R., 1980, Automorphism on ctalan trees and bracketing. Journal of
Combinatorial Theory, B 28, pp. 75–90.
DUCHAINEAU, M.A., WOLINSKY, M. and SIGETI, D.E., 1997, ROAMING terrain: real-time
optimally adapting meshes. In Proceedings of IEEE Visualization’97, 13–24 October
1997, Phoenix, Arizona (IEEE Computer society), pp. 81–88.
ECK, M., DEROSE, T., DUCHAMP, T., HOPPE, H., LOUNSBERY, M. and STUETZLE, W., 1995,
Multiresolution analysis of arbitrary meshes. Computer Graphics, 29, pp. 173–182.
EI-SANA, J. and VARSHNEY, A., 1999, Generalized view-dependent simplification. Computer
Graphics Forum, 18, pp. 84–94.
EVANS, W., KIRKPATRICK, D. and TOWNSEND, G., 1997, Right triangular irregular networks,
Technical Report 97-09, Department of Computer Science (Tucson, Arizona;
University of Arizona).
GARLAND, M., 1999, Quadric based polygon surface simplification. 3–8 August, Los Angeles,
ACM Press. (Pittsburg, PA: School of Computer Science, Carnegie Mellon
University), CMU-CS-99-105.
GARLAND, M. and HECKBERT, P.S., 1997, Surface simplification using quadric error metrics.
In Proceedings of SIGGRAPH’97, 3–8 August, Los Angeles, ACM Press, pp. 209–216.
GERSTNER, T., 2003, Multiresolution compression and visualization of global topographic
data. GeoInformatica, 7, pp. 7–32.
GROSS, M.H., GATTI, R. and STAADT, O.G., 1995, Fast multiresolution surface meshing. In
Proceedings of IEEE Visualization ’95, 29 October–3 November 1995, IEEE
Computer Society Press, Atlanta, Georgia, USA, pp. 135–142, Atlanta, GA.
HECKBERT, P.S. and GARLAND, M., 1997, Survey of polygonal surface simplification algorithms.
Technical Report CMU-CS, Carnegie Mellon University Pittsburgh, Pennsylvania, USA.
HELLER, M., 1990, Triangulation algorithms for adaptive terrain modelling. In Proceedings of
the 4th International Symposium on Spatial Data Handling, IGU Commission on GIS,
Zurich, Switzerland, 23–27 July, pp. 163–174, Zurich.
HOPPE, H., 1996, Progressive meshes. In Proceedings of SIGGRAPH’96, 4–9 August, pp.
99–108, New Orleans, Louisiana, ACM, Press.
HOPPE, H., 1998, Smooth view-dependent level_of_detail control and its application to terrain
rendering. In Proceedings of IEEE Visualization 1998, 18–23 October, IEEE
Computer Park, NC, pp. 35–42.
KIM, J. and LEE, S., 2001, Truly selective refinement of progressive meshes. In Proceedings of
Graphics Interface 2001, 7–9 June, Ottawa, Ontario, Canada, pp. 101–110.
1038 Multi-resolution terrain models
LEE, M. and SAMET, H., 2000, Navigating through triangle meshes implemented as linear
quadtrees. ACM Transactions on Graphics, 19, pp. 79–121.
LINDSTROM, P. and PASCUCCI, V., 2002, Terrain simplification simplified: a general
framework for view-dependent out-of-core visualization. IEEE Transactions on
Visualization and Computer Graphics, 8, pp. 239–254.
LINDSTROM, P., KOLLER, D.F., HODGES, L., RIBARSKY, W., FAUST, N. and TURNER, G.,
1995, Level of detail management for real-time rendering of phototextured terrain.
Available online at: http://www.gvu.gatech.edu/gvu/reports/1995/(accessed 20
Feburary 2004).
LINDSTROM, P., KOLLER, D., RIBARSKY, W.F., HODGES, L., FAUST, N. and TURNER, G.,
1996, Real-time continuous level of detail rendering of height fields. Computer
Graphics, 20, pp. 109–118.
MAHDI, A. and CHRIS, W., 1998, Representation of 3-D elevation in terrain databases using
hierarchical triangulated irregular networks: a comparative analysis. International
Journal of Geographical Information Science, 12, pp. 853–873.
MALLAT, S.G., 1989, A theory for multiresolution signal decomposition: the wavelet
representation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 11,
pp. 674–693.
PARK, D., CHO, H. and KIM, Y., 2001, A TIN compression method using Delaunay
triangulation. International Journal of Geographical Information Science, 15, pp.
255–270.
PEDRINI, H., 2000, An adaptive method for terrain surface approximation based on triangular
meshes. Ph.D. dissertation, Department of Electrical, Computer, and System
Engineering, Rensselaer Polytechnic Institute, Troy-NY, USA.
PUPPO, E., 1998, Variable resolution triangulations. Computational Geometry Theory and
Applications, 11, pp. 219–238.
RIBELLES, J., LOPEZ, A., BELMONTE, O., REMOLAR, I. and CHOVER, M., 2002, Multiresolution
modeling of arbitrary polygonal surfaces: a characterization. Computer Graphics, 26,
pp. 449–462.
ROETTGER, S., HEIDRICH, W., SLUSALLEK, P. and SEIDEL, H.-P., 1998, Real-time generation
of continuous levels of detail for height fields. In Proceedings of WSCG ’98, 9–13
February, Plazen–Bury Czech Republic, pp. 315–322.
SCHROEDER, W., ZARGE, J. and LORENSEN, W., 1992, Decimation of triangle meshes.
Computer Graphics, 26, pp. 65–70.
SHI, W.Z., YANG, B.S. and LI, Q.Q., 2003, An object-oriented data model for complex objects
in three-dimensional Geographical Information Systems. International Journal of
Geographical Information Science, 17, pp. 411–430.
SOUCY, M. and LAURENDEAU, D., 1996, Multiresolution surface modeling based on
hierarchical triangulation. Computer Vision and Image Understanding, 63, pp. 1–14.
VOIGTMANN, A., BECKER, L. and HINRICHS, K., 1997, A hierarchical model for multi-
resolution surface reconstruction. Graphical Models and Image Processing, 59, pp.
333–348.
WU, J. and AMARATUNGA, K., 2003, Wavelet triangulated irregular networks. International
Journal of Geographical Information Science, 17, pp. 273–289.
XIA, J., EL-SANA, J. and VARSHNEY, A., 1997, Adaptive real-time level-of-detail-based
rendering for polygonal model. IEEE Transactions on Visualization and Computer
Graphics, 3, pp. 171–183.
YANG, B.S., LI, Q.Q. and SHI, W.Z., 2005a, Constructing multi-resolution triangulated
irregular network model for visualization. Computers & Geosciences, 31, pp. 77–86.
YANG, B.S., SHI, W.Z. and LI, Q.Q., 2005b, A dynamic method for generating multi-
resolution TIN models. Photogrammetric Engineering &Remote Sensing, 71(8), pp.
917–927.