Building a Large-Scale Synthetic Environment (Part 1)
Rester connecté, inscrivez-vous à notre infolettre.
This article is part 1 in a set of posts describing the challenges of producing large-scale synthetic environments. The second and final part will be published in a few weeks.
One terrain database does not fit every simulation.
You have the high flyers that soar 30k-50k feet above the ground, the lower level ones like helicopter simulators hovering at 5k feet or lower, and then of course tanks or unmounted infantry simulators that roll, run, or walk on the ground, and finally the marine simulators that cover vast oceans – from above and below.
Each of these virtual terrains – or synthetic environments – require a different level of detail with regard to the topology, roads, or buildings in order to satisfy training requirements as well as manage performance issues. So, for example, a fast jet simulator will not have high resolution textures for every single house in a village below that is swishing by at 1,100 mph. What’s the point? However, if an unmounted soldier in a simulation is entering the village, then not only are the building’s shapes and textures important, but its interiors probably are too.
Anyone who has created large-scale databases knows all too well the huge amount of time involved in generating terrain, modeling buildings or full sites, texturing and materializing (if sensors will be used), and adding road networks and pattern of life. Typically, much of this work is performed by hand. But as the need for very large-scale terrains grows, the task of producing, say, several geocells, or a 400km by 400km area requires either a) many more hands, or b) automation.
Presagis has helped pioneer advances in automated terrain generation through both external collaborations and internal development:
- In 2017, Presagis participated in the Foreign Comparative Test (FCT) project that sought to speed the production of high-resolution 3D terrain to support the preparation for tactical operations. Along with Vricon and Luciad, Presagis was able to demonstrate that FCT prototypes could apply non-traditional technologies to produce high-quality terrain databases, with more accurate models, on a vast scale using unclassified satellite imagery.
- Terra Vista has also been a front-runner in the creation of synthetic terrains. Built by Presagis, Terra Vista is a terrain generation software that can convert imagery, elevation, vectors, and other GIS data into optimized 3D simulation-ready virtual environments. But rather than simply convert numbers into vectors, Terra Vista is able to stylize roads and environments to make them match their locale through the use of visual “recipes” that will, for example, make a neighborhood in Kansas look very different than a neighborhood in Kandahar.
Automating Terrain Generation
A few companies are working on automated solutions to produce vast terrains based on Geographic Information System (GIS) data. For the most part, these environments consist of landscapes with rolling hills and countryside. Even fewer companies are tackling the challenge of automatically producing urban areas (at street level). However, those that have keep hitting the same roadblocks:
- Insufficient GIS data
- Corrupted public GIS data
- Uncorrelated GIS data
- Amount of labor required to create complex roads/road networks
In developing VELOCITY, a software solution that automates the production of large simulation environments, Presagis understood that to achieve a realistic urban environment, roads were key. “We started with roads because we believe they are the spine and veins of a city, and that human activities naturally flow from there,” explains Lead Architect Dave Lajoie.
A common starting point for users or organizations is the immense open-source database known as OpenStreetMap, or OSM, a crowd-sourced project aims to create a free, editable map of the world. As with anything crowd-sourced, inaccuracies, errors, or “vandalism” are common – which does not lend well to creating accurate and detailed terrain databases for use in simulations.
So, to avoid the common roadblocks, Presagis, developed a tool within VELOCITY that a) performs a statistical analysis of the entire database for the whole planet; b) extracts only meaningful data that is useable; c) identifies what needs to be fixed; and d) implements the fixes. Lajoie cites an example, “In one instance we kept seeing a road that was sometimes referred to as four lanes, and other times as 2 lanes. This is not helpful if you’re attempting to build a road network -- so we created tools to fix these aspects.”
With clean OSM data, VELOCITY is ready to automatically create roads. With good vectors, generating roads with simple intersections is easy enough. However, things get tricky when reproducing digital versions of complex road intersections, T-junctions with crossings, angled intersections, N-junctions, roundabouts, and so on.
To overcome this challenge, rather than reinvent the wheel, VELOCITY’s modular architecture is able to leverage existing or established solutions to help resolve these types of items. “The beauty of VELOCITY is that it can use existing applications in your organization,” explains Lajoie. “For example, we used a popular solution to tackle complex road networks. By breaking them down, we made them more manageable and allow distributed processing which allows scaling.”
Lots, Cadasters, and Building Footprints
With a road network in place, you are then faced with empty city blocks, or, essentially, the space between roads. These spaces can contain a single house, hundreds of houses, a vacant lot, a shopping mall, an airport, or… anything, really.
VELOCITY is able to leverage a wide variety of data to help close these information gaps whether they be cadastral databases, commercial sources such as Lux Carta, point cloud data gathered from UAVs, or any other type of vector format in existence. The aforementioned sources can not only provide geo-specific building footprints, but also give you accurate building heights. So, once the database is generated, the building’s shape, volume, rooftop, textures and openings can go from being completely geo-specific to a great geo-representative approximation depending on the availability and precision of the input source data.
Once all known buildings have been generated, there is likely to be large, outlying sections of the city for which there is no precise data, such as suburbs, or remote industrial sectors. For these areas, VELOCITY is able to “backfill” the cityscape in a very realistic manner. This very sophisticated procedural generation analyzes the surrounding area and produces buildings, vegetation, and clutter that is not only in keeping with the country, but also with the local area. The placement of buildings, their category, spacing, style, and density are all determined using an advanced series of algorithms and techniques that ensure they mesh as much as possible with reality.
Geo-Specific vs. Geo-Typical
As you acquire more data, the need to “backfill” the environment with procedurally-generated geo-typical buildings diminishes. Of course, these procedural areas are not geo-specific, but geo-typical and would not be used when accurate representations are needed. On the other hand, areas that were generated using precise data for the roads, building footprints, and heights are ideal for when “ground truth” is required.
If the goal is to produce a terrain database for training, the design of the database needs to satisfy training requirements. The balance between geo-specific and geo-typical is determined by what is important to your synthetic environment. As mentioned earlier, an aircraft simulation does not necessarily need high-precision for ground elements. Thus, a procedurally generated neighborhood on the outskirts of a city is acceptable. If a helicopter simulator will be conducting an operation in that neighborhood, then it should ideally be as geo-specific as possible and generated using actual, fresh, and precise data in order to enable accurate and timely decision-making.
Presagis built VELOCITY around these principles in order to ensure that users can maximize their data and produce meaningful content that serves a desired purpose.
So what makes a good simulation environment?
For starters, it’s more than just static terrain. A good simulation environment is a data-rich, high-quality, multi-function database that conforms to your requirements, scales to your increasing level of data, and adapts to your needs.
VELOCITY is able to analyze and transform diverse geospatial data streams to create massive, rich 3D virtual environments for multiple uses. Applications range from the creation of geo-specific terrain databases for mission rehearsal to the design of procedurally generated geo-typical environments for training, to geospatial data processing for the segmentation, extraction and analysis for geo-intelligence application.
From generating terrain and creating road networks, to inserting building footprints and producing structures, we’ve described how VELOCITY can be used to produce large scale realistic geo-specific environments, as well as vast geo-typical spaces.
But it doesn’t stop there.
In Part 2 of Building Large Scale Environments, we will dive into what makes an environment immersive, and walk through how VELOCITY deals with vegetation, textures, and materials for use in sensor-ready simulations such as infrared, thermal, and night vision.
Rester connecté, inscrivez-vous à notre infolettre.
VELOCITY automates the production of large-scale synthetic environments including buildings, vegetation, and pattern of life.