Continuous Adaptive Terrain Modeling
For DIS And Other Simulation Applications

Carl Suttle
Product Manager
MultiGen Inc, Suite 500, 550 Winchester Blvd South
San Jose, California

ABSTRACT

Interoperability across heterogeneous simulators provides challenges for visual simulations and visual databases within those simulators. Advances in visual database modelling capability are required and possible because of increased processing and graphics power of image generators. Old and new modelling capabilities and design constraints discussed in this paper provide insight into how interoperability may become less challenging for OpenFlight format database applications.

KEYWORDS

Distributed Interactive Simulation, Continuous Adaptive Terrain, MultiGen, Silicon Graphics, Morphing, Triangular Irregular Networks, Transportability, Correlation.

INTRODUCTION

Real time image generation is today a major component of man in the loop simulators without which many training and research exercises would not be achievable. Exercises undertaken using Distributed Interactive Simulation (DIS) can rely heavily on the real time visual feedback which enables simulator operators to control the simulator and make tactical decisions.

Performance limitations of even the most advanced image generators (IGs) force the use of visual scene simplification techniques and specific equipment tuning for convincing real time scenes to be generated.

Because of this providers of networked simulators cannot guarantee that simulator operators will see consistent visual scenes. This may give one operator tactical advantage over another. This "non fair fight" reduces the effectiveness of DIS.

For networked simulators of the same type performing exercises in one domain, ground warfare simulation for instance, these simplifications are consistent across simulators. If the IGs in the system are not overloaded and the visual scene is presented with consistent scene quality, any operations between the participants can be "fair".

For heterogeneous simulation networks, as allowed by DIS, IG features, performance and therefore real-time visual representation of a scene will vary from simulator to simulator. However visual feedback given to the simulator operators can be correlated to some extent. If correlation is within limits defined by the operational requirement the fight can be fair.

Illustrated in table 1 are four types of DIS node and their typical requirements for operator feedback. Heavy reliance is placed on the Synthetic Environment database which, as a subset, contains the visual database.

System Type	Feedback Requirements	Information Source	Display Examples
Man in the loop simulators	Operators need to sense their environment in a similar manner to that presented in the equipment being simulated	DIS PDUs and Synthetic Environment databases (OpenFlight, proprietary databases)	Radar, Sonar, IR, Audio, Visual and other sensor simulations displaying external scenario and environment elements
Computer Generated Forces (CGF)	CGF systems need to sense the environment from computer databases	DIS PDUs and Synthetic Environment databases (ctdb, mrctdb)	A 2 D representation of the scenario
Real Systems	Stimulation of on board systems to inject simulated targets	Systems stimulated by DIS PDUs	Radar, Sonar, IR, Audio, Visual and other sensor simulations displaying a mix of real and simulated data
Stealth Systems	Situation Display	DIS PDUs and Synthetic Environment databases	A 3D representation of the scenario

Table 1. DIS Node Types and Feedback Requirements

The correlation of visual scenes is largely controlled by the visual database. Fair fight problems, time and cost might be reduced if databases were more transportable across different types of image generator and re-usable across domain applications. Two goals for visual databases in DIS are then:

• Correlation
• Transportability

MultiGen Inc. have been developing new visual database terrain generation tools which build transportable databases for OpenFlight format users. Using these databases the level of correlation and performance of individual image generators can be selected on line rather than being hard coded into the visual database.

This paper details current visual database design trade-offs for OpenFlight visual databases and how these trade-offs will be reduced and database transportability enhanced in the future.

THE VISUAL SYSTEM AND IMAGE MANAGEMENT

Referring to figure 1, a visual system may be considered as three components, a visual database, IG, and a display system. The visual database contains data defining geometry, colour and textures which represent the simulated environment. The IG renders the information stored in the database from a viewpoint defined by the simulator. The display system presents the information to the operator in the correct format using a projected, CRT or headmounted display.

An IG cannot process in real time all the data in the database. It is the function of overload management to ensure the IG only draws what is necessary, reducing visual detail as required to maintain the real time operation and scene iteration rates required for effective man in the loop simulation.

IG manufacturers use similar techniques to implement overload management. Viewing frustum culling, level of detail (LOD) switching, instancing, object priority and update rate feedback and control are used in most IGs.

The rules and data that enable overload management are defined within the visual database.

CURRENT VISUAL DATABASE DESIGN - VIEWING FRUSTUM CULLING GROUP SIZE AND LEVEL OF DETAIL FOR OPTIMAL PERFORMANCE.

5.1. Culling. For visual systems culling is the process of rejecting data which does not need to be processed because it is outside of the simulator operators viewing frustum. OpenFlight format and most traditional terrain databases use rectangular areas for culling group definition. In this implementation culling and LOD geometry is fundamentally different to the geometry to be rendered. In fig 2 any shaded area outside of the simulator operators viewing frustum represents information that the visual scene will process unnecessarily.

For efficient operation the database designer must optimise the time taken culling. Too many groups and the culling process takes too long, reducing the frame rate. Too few and the IG will process unnecessary data, reducing visual quality inside the viewing frustum or reducing the frame rate.

Because culling group sizes are regular and based on terrain surface area, the number of groups in the viewing frustum will depend on the range of the eyepoint from the culling groups and the definition of the viewing frustum. Unfortunately these parameters may change within a simulation exercise, one group size cannot be efficient for all applications. For instance for optimal performance databases designed for high flight level (altitude) applications will have much bigger group sizes than those for ground warfare.

5.2 Level Of Detail. Level of Detail Switching can be used to change culling group size with respect to eyepoint distance. See fig 3. Also as the content of the coarse LOD groups is never seen at close range they may contain a coarser representation of the visual scene than the fine LODs. Fewer polygons at large distances allows more polygons at closer ranges where they are more effective.

Group culling and LOD form the basis for overload control in most IGs, however LOD itself requires cull processing time and introduces undesirable dynamic and static artefacts into the visual scene which require careful management.

5.3. Dynamic Artefacts Are Introduced As Levels Of Detail Switch. LOD switches in and out different accuracy representations of the terrain. The database designer must balance the effects of these switches such that image generator performance is maintained whilst distracting visual "pops" are kept to a minimum. Fade LOD can be used to mask switching artefacts by drawing and blending two levels of detail at the same time. This is a good technique for reducing visual artefacts but increases the load on the image generator at the worst possible time.

5.4. Static Artefacts Are Introduced Into The Visual Scene. These are produced by requirements to match vertices at LOD and culling group boundaries as shown in fig 4. Part of the database modeller task is to reserve some of the available vertices along group edges whilst maintaining enough interior vertices for database accuracy. Too few vertices along edges results in unsightly "cliffs" at group boundaries. LOD groups A and B of figure 4 are compared in table 2.

Group	No of Polygons	Interior Vertices	Exterior Vertices	Comments
A	33	10	15	Maintains good boundary detail but has reduced interior accuracy. Higher number of exterior vertices limits the minimum number of vertices in the coarse LOD.
B	32	22	7	Interior accuracy is increased and adjoining coarse LOD can have fewer vertices but long thin (and unsightly) polygons are produced at group boundaries.
C	28			For maintenance of a continuous terrain skin proportionally more vertices of the coarse LOD have to be used at the Cs exterior to interface to group A than group B

Table 2. Comparison of LOD Groups A and B

Using these load management techniques visual system performance constraints determine that:

• Visual database modelling is a time consuming, skilled and specialised task.
• Visual databases are highly tuned to the target image generator performance
• Visual databases are highly tuned to the target application.

5.5. Comparing The Performance Of Visual Systems Using Current Visual Database Design And DIS Requirements.

• Correlation - Real time synchronisation and operation at acceptable fidelity levels can be achieved but using LOD results in unpredictable dynamic correlation errors. If this is the case then correlation between DIS nodes is also unpredictable.
  
• Transportability - The cost associated with visual database construction is very high. It is desirable for cost, time and correlation reasons that visual databases be transportable to IGs of different performance and manufacturer. Unfortunately todayÆs databases are not easily transportable because:
  

  • Many databases are constructed in proprietary or highly specialised formats.
  • Databases are tuned for one type of image generator. There may be many man weeks of effort required to compensate for changes in IG architecture.
  • Databases are tuned to one type of training task. Converted databases will still only be useful for that task. There is no standard way of constructing entity or terrain databases.

CORRELATION AND TRANSPORTABILITY ENHANCEMENTS

The following describes techniques investigated and in implementation at MultiGen Inc. and Silicon Graphics Inc. (SGI) . These techniques are not intended to cure system design problems of either DIS or IGs but to provide better transportability and re use of visual databases across DIS and other visual simulation applications. Please Note: The following comments only apply to OpenFlight based systems, OpenFlight is a visual simulation interchange format that has limitations for CGF applications, OpenFlight currently provides only limited topology and topography data.

6.1. Transportability, Formats and APIs. As has been described transportability of visual databases is more than a format problem, even so database conversion is required. Examples of successful database formats are:

Evans and Sutherland, GDF
MultiGen Inc., OpenFlight
Coryphaeus, DWB
Lockheed Martin, Target
Loral, S1000

MultiGen database modelling tools can be used to convert between all of these formats and MultiGenÆs OpenFlight format is by far the most widely used, Evans and Sutherland being the most recent adopters.

The introduction of APIs to provide easier format interchange is a popular subject in the industry. APIs are available for S1000, OpenFlight and other database formats.

The SEDRIS API initiative is especially interesting. Sponsored by STRICOM, SEDRIS contains a superset of all data and structures required for interoperable simulation databases. The data definition of SEDRIS is much more complete than any of the above formats or APIs. MultiGen Inc. look forward to its implementation and acceptance by the simulation community.

6.2. CAT Database Mission And IG Transportability. The following describes Continuous Adaptive Terrain(CAT) generation and real-time image generation.

6.3. Culling, Level of Detail and Morphing. The rectangular design of terrain grouping for culling forces compromises on the databases designer. Even with a well designed database culling group size is always a compromise. Work completed at MultiGen by Lee Willis implements triangular culling and LOD groups within a recursive Triangular Irregular Network (TIN) to provide a Continuous Adaptive Terrain. Within a CAT database the culling and LOD boundaries are also the geometry rendered by the IG.

As can be seen in fig. 5 each triangle recursively subdivides to create finer LODs until the required visual fidelity is produced. Triangle position and size is irregular and determined at each LOD by terrain surface fit algorithms. At database creation time morphing information is stored with the database such that continuous morphing between LODs is available.

6.4. CAT has several advantages over traditional rectangular TIN systems. 

6.4.1. Lack of Dynamic LOD Switching Artefacts. When the LOD switching parameter instructs a switch from course to fine representations the course triangle is replaced by into 2, 3 or 4 triangles with the same outline, coplanar with the original triangle (fig 5 a and b). At this time the simulator user will not be able to determine an LOD switch has taken place. As the LOD switching parameter moves towards the next switch the new triangles morph towards their final position. Morphing is completed at the same time as the next switch (fig 5. c). Using CAT dynamic switching artefacts of LOD are replaced by morphing artefacts, these are less noticeable in all operational conditions. 

6.4.2. Static Artefacts. As shown in fig 5 LOD and culling group boundaries are non regular and edges can morph outside of their original positions. This eliminates the static artefact problems associated with rectangular groups. The database modeler no longer has to worry about interior and exterior vertex proportions. In fig 5 the thick edge is a ridge line which continually morphs to an accurate representation over 4 LODs, Figure 6 shows coarse and dual LOD examples illustrating the lack of visual artefacts and efficient culling and Fig 7 shows screen shots from a CAT application. 

With CAT, the database designer does not have to decide culling group sizes, these being the same size as displayed terrain triangles. Recursive triangular culling and LOD groups allow one database to be used across different training scenarios. 

6.5. Image Generator Performance And Real Time Control Of CAT. It is rare in real time visual simulation that improvements in visual quality are free, and this is not the case here. The CAT culling process requires more computing power and the morphing process has to be performed. Fortunately these are CPU processes and can be multiprocessed. 

6.6. LOD Control May Be Modified By IG Or External Control. CAT allows the IG to control various aspects at runtime such that the database can be tuned on line for IG performance and mission profile. 

6.6.1. Dynamic Control Of Image Generator Load. The IG can smoothly reduce LOD anywhere in the viewing frustum to maintain real time. If the same database is loaded on IGs of different power the user will be able to control whether they show the same visual scene at different update rates or the same update rates at different scene complexity. If the same scene and update rate are required then the more powerful IG can be detuned appropriately. 

6.6.2. Dynamic Control Of LOD. LOD variations can control the processing of the database such that the visual scene is optimised for different mission profiles. Figure 8 shows two applications running the same database with each application tuning the visual scene. The medium level flight simulator has a linear reduction in polygonal density v range and a conservative polygon density for its fine LOD. The ground warfare simulator has a much higher density for the first 5 km dropping off rapidly to a lower LOD. As the aircraft in the first example reduces its altitude for landing the LOD processing may transition to settings similar to those for ground warfare, these being more appropriate for landing. 

6.7. Real Time Control Over The Balance And Rate Of Cull And Morph. At any time during a simulation exercise the situation may be such that the computing time taken to cull, morph and draw the database may not be equal. This is the optimal loading for IGs where the cull, morph and draw are parallel processes. The simulation application may control the cull /morph balance. It is expected that cull/morph processes will be under rate control , in applications where the eyepoint moves relatively slowly cull and morph processing every frame will not be required. Draw time is a function of the LOD drawn, this is also under real time software control. 

6.8. Summary Of CAT Attributes.

• The terrain database can be tuned for the application and (OpenFlight compatible) IG power at runtime instead of being hard coded in the visual database.
• Morphing has less visual artifacts than LOD switching.
• No LOD or Group boundary artefacts.
• Less database design required.
• Cull /Morph/Draw balance control.
• More runtime CPU power and database creation pre processing required.
• Only useful on IGs that support CAT components of OpenFlight format.

Conclusions - Comparing The Performance Of Visual Systems Using CAT And DIS Requirements

CAT is not the answer to DIS or IG system design problems. IGs running CAT still use the same overload management techniques, but these are no longer hard coded into the visual database. Database design is easier, production takes less time and databases are more re-usable, reducing cost.

• Correlation - Real time synchronisation and operation at acceptable fidelity levels can be achieved. LOD parameters may be controlled on line until visual scenes are of consistent quality across IGs of different power.
  
• Transportability - CAT OpenFlight databases are transportable and cost can be reduced.
  

  • OpenFlight is a public domain format. Read and Write APIs are available.
  • Databases can be tuned on line for all image generators capable of running CAT data structures.
  • Databases can be tuned on line for the various stages of a training task.

 

---

MultiGen is a registered trademark and GameGen and SmartScene are trademarks of MultiGen, Inc. All other trademarks mentioned herein are property of their respective companies.

Copyright 1997 MultiGen Inc.