CREAS-MAP PROJECT: 3D reconstruction from satellite images is helping create simulations faster

Our AI-driven approach to 3D reconstruction from satellite images helps analysts, planners, and engineers create simulations of the real world faster than ever before.

Key takeaways: 

• Traditional techniques for creating 3D simulations are manual, expensive, and slow
• We have developed an automated approach to 3D reconstruction from satellite images that is fast, efficient, and accurate
• Learn how our satellite 3D modeling techniques work
• Find out how it can be used in your industry 

Accurate 3D simulations of the real world can make a major difference to many organizations’ operations. They let planners and analysts explore a site of interest, identify risks, and understand layout in a far more meaningful, engaging way. Compare the experience of studying a 2D map with wearing a VR headset and ‘walking’ around a 3D simulation of a place.  

At LuxCarta, we recently collaborated with CS-Group and Inria - France’s national IT research institute - on a project to address this issue. Named CREAS-MAP (Classification, Re-construction and Automated Generation of Mixed Simulation Environments Assisted by Deep Learning), the project is founded by the French DGA (Délégation Generale à l’Armement) in the frame of RAPID funding . It combined remote sensing and 3D mapping to rapidly create realistic simulations for anywhere on Earth. 

We have now integrated this process into our BrightEarth platform - which means all our customers can use it to create 3D simulations. Let’s learn more about this innovative approach to 3D reconstruction from satellite images. 

The need for a faster way to create 3D simulations of real-world places

For many years, the process of creating 3D simulation environments of real-world places was time-consuming and resource-intensive. Simulations were typically created offline and manually by experienced professionals. They used a wide combination of inputs, including advanced satellite imaging, on-the-ground photos, LiDAR, and site visits to create highly accurate versions of real-world places. 

While this process produces very detailed and accurate simulations, it is not always ideal:

• Simulations go ‘out of date’: Simulations quickly go out of date. New buildings or entire neighborhoods can emerge in months, forest fires can denude a landscape, conflicts can radically alter land use. For any time-sensitive scenarios, these factors can render simulations irrelevant. 

• Accuracy over speed: Traditional simulations prioritize the most truthful likeness possible. Although accuracy should always be the goal, ‘perfection’ is often unnecessary. Planners normally need a close representation of buildings, roads or trees in a place - but it doesn’t need to look ‘picture perfect’. 

• Lack of data: In many parts of the world, there is a lack of reliable LiDAR or photographic data about a place, which makes it almost impossible to create 3D simulations. 
  
  

A new approach to 3D reconstruction from satellite images

To address these problems, LuxCarta collaborated with our partners on the CREAS-MAP project. The aim of CREAS-MAP was to explore the potential of using artificial intelligence (AI) in satellite imagery processing to generate 3D scenes that could be streamed to simulation software. Rather than using site visits or traditional survey, and so on, the goal of the project was to create accurate simulations using mainly satellite images and occasionally opensource database like OSM.

The process we developed follows several steps:

1. A raw dataset of satellite imagery of anywhere on the Earth’s surface is selected.
2. Buildings, trees, and roads are segmented using deep learning and vectorized by computational geometry
3. Calibration model adjusts raw images according to a reference. 
4. We estimate digital surface models, digital terrain models and digital elevation models from satellite imagery.
5. The images then undergo a 2D reconstruction process. 
6. We then use an ortho-imaging process to extrude buildings and trees from the landscape and create a 3D reconstruction from satellite images.
7. Our AI texturing process adds realistic color, detail and roof shapes to the 3D environment.
8. The environment is then available for massive online streaming in common simulation software (OGC formats include WMS, WFS, CDB and 3D Tiles) or for offline software 

In-depth: Download our paper from NATO’s Science & Technology Organization website for a complete description of the process.  

Benefits of 3D satellite imagery simulations

Our approach provides several key advantages to any organization that requires simulations of the real world.

• Speed: Our method for geospatial data visualization is very fast. For example, we ran a test on Mourmelon-le-Grand, a small town in northern France (0.7 km2, with 2,377 buildings). The pipeline used a single satellite image and took just a few minutes to produce an accurate 3D simulation environment. We tested again for greater accuracy by using a stereo pair and street-level images (for 318 facades) for greater fidelity. The run time was just 50 minutes. 

• Accuracy: As part of our testing, we used a Microsoft HoloLens to validate the positioning and texturing of buildings and texturization. A member of our consortium wore the HoloLens headset that streamed our 3D reconstruction of Mourmelon-le-Grand, while walking around the real town. The demonstration showed that our simulation recreated the town very accurately (of course, some of the architectural details or colorings were not exactly the same). 

• Reduced input: Our satellite 3D modeling technique means that customers can create realistic simulations of real-world places with nothing more than a single satellite image. 

• Up to date: One of the greatest benefits of 3D satellite imagery for simulations is timeliness. It is possible to use the very latest satellite images of a place and create a simulation environment. This means that organizations no longer rely on out-of-date simulations that could be inaccurate. In theory, it would be possible to create a simulation from a satellite image taken just a few minutes ago (under ideal weather conditions). 
  
  

How 3D reconstruction from satellite images helps your sector

The creation of geospatial data visualization for 3D simulation can support a wide range of sectors, including:

• Defense and intelligence: Plan military operations based on the most up-to-date information about battlefield. BrightEarth’s remote sensing and 3D mapping engine creates a reliable representation of the real-world environment. It can pick up crucial details - from new buildings and roads, to changing forest, and stream these into your simulation and training platforms. 

• Engineering and architecture: Place your CAD and BIM models in a 3D geospatial visualization. Experience what a new building or piece of infrastructure will look like in the landscape before you build it. 

• Public services: 3D reconstruction from satellite images helps a variety of government services understand their territory better. For housing ministries, it can help identify informal settlements. For tax authorities, it can be used to analyze land use (e.g. discover construction of ‘hidden’ property on private land). For planning departments, it can be used to inform decisions about infrastructure, for transport departments, it can inform new routes - and so on. 
• Training: Our satellite 3D modeling techniques support simulations for a wide variety of training scenarios. For emergency responders, it supports the creation of realistic ‘replicas’ of your city when training police, ambulance or fire engine drivers. For logistics firms, it means they can train lorry drivers. For drone companies, it lets them train operators in places that look and feel like the real world. 

In action: How the French navy benefits from BrightEarth’s 3D mapping

Try 3D reconstruction from satellite images for yourself

We are pleased to announce that the technology we developed for the CREAS-MAPS project is now publicly available in BrightEarth. This means that all of our customers can experience the benefits of 3D satellite imagery by streaming massive geodata in major simulation Software but also on Cesium Compatible Engine like Unreal Engine, Unity, or NVIDIA Omniverse. 

Whether you want to create simulations for defense purposes, research, training, planning, entertainment, gaming - or anything else - BrightEarth allows you to generate realistic 3D reconstructions from anywhere on Earth and stream these to your simulation environment of choice. 

Contact us to start using BrightEarth, and try it for yourself today.