79,000 sqkm. 95,000 Nadir Images. One 3D Map.

Posted on August 11, 2017 by Ladislav Horky

In a joint effort with Mapy.cz, Melown Technologies has carried out arguably one of the largest 3D mapping projects ever done, covering the entire Czech Republic with a photogrammetric 3D map created from 12.5cm/px nadir imagery, with select portions based on 10cm/px oblique imagery.

In the beginning we were facing the task of turning about 95,000 nadir images covering 79,000 square kilometers at 60/30 overlap and 12.5cm/px resolution into one large 3D model. Then we needed to process some 313,000 oblique images covering over 5,150 square kilometers at 60/40 nadir overlap and 10cm/px resolution, to create photogrammetric 3D models of all but the smallest towns in the Czech Republic. And finally, we needed to fuse all this data into a single interactive country-wide photogrammetric 3D model.

Area covered by country and city models

We approached this task using our in-house photogrammetry system, codenamed Vadstena, together with our open source VTS 3D Geospatial Software Stack.

Starting with the bulk of the nadir imagery we decided, mainly for manageability and shipping reasons, to split the computation into a regular grid with each square cell being about 15 kilometers wide. This way we had obtained about 350 “targets” that were then computed independently. This also proved practical, since we did not have the luxury of having all the data available at the same time. Splitting the monstrous 18 TB dataset into targets allowed us to have a substantial portion of the dataset processed while the aerial survey was still underway. For the remaining oblique city datasets, we let our algorithm split the computation of each city by itself.

Both Vadstena and the VTS Stack scaled well to the task, both by design and by implementation. The pipelined, parallel nature of Vadstena allows it to run on multiple nodes against one database and an NAS. Nodes can be removed or added independently without interrupting the computation. On the output side, the VTS Stack sports a size-agnostic data model and powerful, scalable backend components for serving data.

For successful deployment on Mapy.cz, we needed to fuse several terabytes of 3D models. The ultimate goal, of course, was one seamless detailed 3D model of the country. Yet experience taught us that operations like adding, updating or removing models need to be low-risk, reversible, no-downtime operations and should preferably involve moving only the least possible amounts of data. VTS handles this gracefully: it can fuse models in a hierarchical manner, first combining logically related models into one which is then fused with the rest. Even though we usually need to display many models as a single surface, VTS treats all the models as layers that can be turned on and off even on the client side. See the docs for the full VTS Stack documentation.

Thanks to deployment on Mapy.cz with its 750,000 daily visitors, VTS 3D Geospatial Software Stack may well have become the most widely used open source 3D rendering solution in the world.

The result of the project, complete with geocoding and route planning, is available to the general public at Mapy.cz website (Czech language only).