Providing a photorealistic, geo-typical, geo-referenced, high-performant and updated, 3D digital terrain model of the entire globe has been impossible until now. The reasons are manifold: any manual approach is much too slow, automated methods are lacking regional specific aspects, the amount of satellite input imagery data and necessary processing power for automated methods is just too massive or client-side rendering of photogrammetric models is slow.
For a flight simulation computer game introduced in 2020 our team developed a massively scalable, end-to-end cloud-computing based geospatial platform extracting terrain and infrastructure attributes from a current global satellite imagery data set and transforming this into a WGS84 georeferenced, geo-typical, photo-realistic, and semantically defined 3D digital twin of earth’s infrastructure, capable of being streamed or stored entirely offline.
The technology consists of advanced AI/ML computer vision algorithms and geospatial analysis tools that process satellite and aerial imagery (including RGB, Infrared and LiDAR point clouds) to detect and extract features in those images: terrain, biomes, buildings, roads, vegetation, water bodies, and more. This architecture enabled to process building footprints for the whole inhabited surface of the globe every 72 hours for updates (approximately 11,600 km2 / minute).
To deliver the accurate reconstruction of approximately 1.4 billion buildings and structures covering the surface of the planet (~ 140 million km2) we developed a highly parallelized and resilient cloud architecture that decomposes multiple petabytes of imagery into tiles and processes them across hundreds of CPU and GPU compute nodes in the cloud. The procedural definition of building defines every single building in a unique, geo-typical, photorealistic, yet highly performant way.
Currently the platform is getting adapted for professional image generators, flight simulators and cases such a pilot training and sensor simulation. This brings addition challenges like persistent, synchronized, multi-view, multi-client states or material attributes at global scale.
Keywords
AI,ANALYTICS,CLOUD COMPUTING,CONTENT GENERATION,ENVIRONMENTS,FLIGHT SIMULATION,IMAGE GENERATOR,TERRAIN
Additional Keywords
geospatial, procedural