Aerial photos are fodder for market research: revealing buildings built, trees logged, wells drilled, livestock and wildlife growth, vehicle movement and much more. Until recently, turning images into intelligence required expensive, expert knowledge – then Lausanne-based start-up Picterra introduced its ‘search engine for the earth’s surface’.
Counting from above can be an essential part of business. A hedge fund wants to verify how many cars are being manufactured in China. An emergency-relief supplier needs to number the tents in refugee camps. A grain trader aims to see which ships are loading or unloading at what locations. Fortunately for them, photos from above – from satellites and drones – are more prevalent than ever. These days, who hasn’t used Google Maps or one of its many competitors offering aerial views? But photos for consumers are not up-to-date or detailed enough for specialist business use.
Say hello to Geospatial Analytics
This sector – also called Geographic Information Systems (GIS) – is the first port of call for companies counting from above. Computerised GIS has been around since the 1960s, and in some applications (say road navigation: the TomTom in your car) it has become completely automated. For counting, it still requires significant human input. “Teaching a conventional GIS system to count objects,” explains Picterra’s CEO, Pierrick Poulenas, “can be very time consuming and expensive.” To get it right, a high-end counter needs to examine some 15 000 examples of its target object, and even then, a data-scientist needs to spend 2-3 weeks fine-tuning, often calling in the help of domain experts. “These machines are not really intelligent,” Poulenas observes, “they just practice a lot.”
Good enough is better than best
Picterra’s answer to cutting time and cost out of conventional GIS is, as Poulenas puts it, “to democratise” the technology. Its object-detection/counting software runs on a central platform that accumulates the knowledge of all its users – now totalling 1500 people in half the world’s countries. By pooling users’ data, the software avoids re-inventing the wheel, i.e. it can skip most of the examples and fine-tuning. Scanning ten snaps of the target and spending a few minutes for initial calibration are all it needs to be up and running. Domain specialists are not needed: their role is taken by the Picterra user community. At the same time, Poulenas concedes, Picterra’s initial results will be slightly less accurate than those of a tailor-made artificial intelligence model. “Our method sacrifices 2-3 per cent in accuracy for much faster, much cheaper results. And that 2-3 per cent can be compensated by users in real time,” in other words, they can correct mistakes and teach the software not to make them again.
As more users deliver more data, Picterra’s software gets smarter and smarter. It learns how to distinguish not just buildings, for instance, but roofs made of copper from those made of tiles. Users continually train the software to reinforce its successes and correct its failures. Critically, users need not be data-scientists or mathematicians: they are business people simply trying to find something or another.
Google the earth
Ultimately what this leads to, Poulenas predicts, is a new kind of whole Earth catalogue. How many Criollo steers in Argentina? Just Picterra it. What’s the inventory of major automakers in their open parking lots? Just Picterra it. If that sounds far-fetched, consider as Poulenas has that “within the next five years, images from satellites and drones will be commoditized, cheap. High-resolution, real-time images will not just be available, they’ll be all over the place.” Picterra’s niche will be in categorising them. So the biggest challenge for the young company is making potential users aware of geospatial analytics’ potential. “How many trees were cut in Amazon today? What’s the traffic like now in Cairo? People need to know these questions can be answered,” says Poulenas. “We want to make the earth’s surface searchable, in real time.”
Plan B won out
Not that this was Picterra’s initial goal. When Poulenas and co-founder Frank de Morsier started three years ago, their aim was to extract geospatial information for electric utilities companies, tracking vegetation or newly built buildings nearby the powerlines. Two things led them to change focus. First, the willingness of their target customers to adapt GIS was slow in coming. Second, the founders realised that detecting and counting things – vegetation, buildings, anything really – is a similar process, regardless of what is being localized. “We realised the potential of our idea,” notes Poulenas, “now we want enough users to realise it as well.”
This article is a part of the ’Millennial CEOs’ series in which we zoom into Millennials’ lives and ask ’What drives the next generation of business leaders?’.