2016 – 2020 ScanVan (FNS)

ScanVan: A distributed 3D digitalization platform for cities

Sion 3D model

3D model of the city of Sion in 2014 – 3D model computed at the DHLAB using aerial images provided by the city of Sion

 

The massive and systematic 3D digitisation of cities is one of the major ICT challenges of the upcoming years. Indeed, the production of reliable 3D urban models is a prerequisite towards the development of autonomous vehicles, urban drones, and more generally the numerical management of cities. In return, these vehicles will be capable of updating the 3D models in real time, enabling the establishment of continuous 4D urban models.

If scanning a building or a small neighbourhood using photogrammetry is now state of the art, no current solution scale well at the level of an entire city. One of the major challenges in digitising a city resides in the sheer quantity of data to be acquired and the processing power required to transform those images into valid 3D models.

This project is an attempt to progress in this direction by developing a vehicle, the ScanVan, capable of scanning cities. By deploying several of those vehicles in the streets, high quality models could be produced in real time. The ScanVan is equipped with a spherical camera (4π) capable of turning surrounding environments into point clouds while driving. This vehicle possesses its own computation system aboard: a dedicated cluster relying on field-programmable gate arrays (FPGAs) and other computational resources especially conceived and optimised for photogrammetric computation. Thus, the vehicle is able to deal with a complete workflow starting with the acquisition of images and ending with the calculation of 3D point models. The advantages of resorting to a distributed computing system such as this one are the following:

 

  • The amount of data to be transmitted is reduced, since all the first steps of the computation are done on-board.
  • The acceleration of the calculations is done in hardware. A specific computational solution, quasi-optimal from an algorithm point of view and very efficient from an energy point of view is achieved using clustered FPGAs.
  • Parts of the calculations are decentralized. Calculations are performed where and when they are needed using mobile systems, preventing the transfer of large amounts of data. This also prevents the need for a complex static centralised computing infrastructure. For this reason, the system is intrinsically scalable.

 

Once the 3D models have been computed in the different ScanVans, they are saved along with the original raw data in mobile hard drives. The content of these hard drives is later downloaded into a docking” station responsible for the uploading of the models as well as their alignment with previously gathered data.

This project aims at building a first prototype of a ScanVan, focusing in particular on the hardware-software pipeline linking the spherical camera with an on-board computation system. As test case, the project ambitions to produce a fully aligned 4D urban model of the city of Sion (Valais, Switzerland), using the ScanVan prototype. The dataset will be publicly available in order to explore potential new services that could benefit from it. In particular, early contact will be made with autonomous vehicles projects currently undergoing in the city.

People:
Nils Hamel / Scientist, EPFL
Vincent Buntinx / Scientist, EPFL
Frédéric Kaplan / Professor, EPFL

Project partners:
Charles Papon / Scientist, HES-SO Valais
Marcelo Kaihara / Scientist, HES-SO Valais
Pierre-André Mudry / Professor, HES-SO Valais