- Title: Smart algorithm builds 'living' 3D model of Zurich
- Date: 1st June 2017
- Summary: ZURICH, SWITZERLAND (RECENT - APRIL, 2017) (REUTERS) VARCITY DEVELOPERS HAYKO RIEMENSCHNEIDER (RIGHT) AND KENNETH VANHOEY (LEFT) LOOKING AT COMPUTER SCREEN VARIOUS OF RIEMENSCHNEIDER AND VANHOEY LOOKING AT COMPUTER SCREEN (SOUNDBITE) (English) HAYKO RIEMENSCHNEIDER, POST DOCTORAL RESEARCH AT ETH ZURICH / VARCITY PROJECT MANAGER, SAYING: "Over a couple of years we collected various amounts of data; aerial data, drone data, static webcams, and publicly available Flickr images. And we've fused all of this information into one coherent 3D model, and we can add new and additional images all the time and create more enriched models of it."
- Embargoed: 15th June 2017 10:53
- Keywords: 3D model algorithm Google Street View ETH Zurich VarCity
- Location: ZURICH, SWITZERLAND
- City: ZURICH, SWITZERLAND
- Country: Switzerland
- Topics: Information Technologies / Computer Sciences,Science
- Reuters ID: LVA0026JG54YJ
- Aspect Ratio: 16:9
- Story Text: Millions of images and video sourced from the internet have been combined by computer scientists to build a rich and detailed three-dimensional model of the city of Zurich.
The platform, known as "VarCity", uses machine learning algorithms to scour the internet for fresh images, evaluate them and automatically update in real-time; in effect creating a living, evolving 3D model of the city.
In practice it resembles the virtual world of a video game, with users able to navigate through a 3D representation of Zurich's historic centre.
"We've built a 3D model of Zurich city by exploiting 4 terabytes of images - that's 3 million images. By using the knowledge we automatically deduce in this project from this data we can build a compressed model that can then be used to walk through in real time just like in a virtual game," said Kenneth Vanhoey, a post doctoral researcher from ETH Zurich.
While programs like Google Street View give users a similar experience, VarCity's developers say their platform is enriched with much more useful data and can be constantly updated.
"We're trying to make a dynamic version which is augmented with knowledge, so that we actually have a 3D model behind it which we can query for knowledge and visualise the knowledge directly," VarCity project manager Hayko Riemenschneider told Reuters.
"We use artificial intelligence to understand every part of the building, so we know where windows, doors and balconies are, and where pedestrians and cars are. And we can augment the city with this additional knowledge," he added.
The computer algorithms for VarCity were developed over five years; a time-span necessary because of the large volumes of data evaluated. The team realised that for the platform to be truly useful it would have to be able to cope with an ever-increasing amount of data.
To create accurate representations of static buildings, as well as moving vehicles and people, the scientists use triangulation to create a three-dimensional model from image data.
"We try to find correspondences between pixels and images. With these corresponding pixels we use a method called triangulation to estimate the depth of the 3D model so that we can create point-clouds - individual points in 3D space - and represent the 3D model," said Riemenschneider.
The tools they've created are scalable and efficient enough to build surface models for the entire city in a couple of hours, rather than what previously took a couple of weeks, added Riemenschneider.
VarCity's real-time evolution could improve the city's infrastructure on a day-to-day basis, the team says. For example, it could recognise where and when pedestrians are on the move and when parking spaces become free.
The researchers have founded several spin-off companies that utilize some of VarCity's features, including a program that provides virtual damage analyses on buildings for insurance companies, for example after extreme weather events.
Another spin-off, called Parquery, was trialled in the Swiss city of Locarno. It evaluates the data gathered from cameras installed near parking spaces, so it knows which parking spaces are free and also whether a car has been mistakenly parked across two spaces.
- Copyright Holder: REUTERS
- Copyright Notice: (c) Copyright Thomson Reuters 2017. Open For Restrictions - http://about.reuters.com/fulllegal.asp
- Usage Terms/Restrictions: None