- If you’re happy then we’ll know it. Realeyes crunches data on people’s facial movements to analyze reactions.
- Realeyes
A London company that says it can track people’s emotions by watching them through their webcams or smartphones has received a €3.6 million ($3.9 million) boost from the European Commission.
The company, Realeyes, is sharing the grant with researchers at Imperial College London, University of Passau in Germany and British gaming company PlayGen, in an attempt to dive deeper into measuring people’s emotions and figure out whether they like what they are seeing.
“This is absolutely a new frontier,” CEO Mihkel Jäätma said. “It’s really taking emotional measurement to another level.”
Realeyes is one of a number of startups around the globe that crunches data on people’s facial movements to analyze reactions. Supporters say the technology could be used to increase driver safety, improve classroom learning or as a type of lie-detector test by police, though the work raises privacy concerns.
Realeyes, founded in 2007, has built up a database of more than five million frames of people’s faces. The bulk of the company’s 40 employees in Budapest and the U.K. are software engineers and scientists with specializations in image processing, artificial intelligence and other computational sciences. The files have up to seven annotations per frame that can say, for example, whether an eyebrow movement indicates surprise or confusion.
“If you’re smiling, we can detect that,” said Jäätma, who grew up in Estonia. “If you’re frowning, that indicates confusion. If you raise your eyebrows, that’s surprise. But of course there are other signals from other parts of the face which give more solid readings.”
Clients, such as Phillips, Procter & Gamble and AOL , use insights from the technology to shape the content in their video advertising, then target viewers by geography or demographics.
“We can get richer data sets to really improve what type of content we make for our audiences, and find out what are the passion points that make them want to share content,” said Mark Melling, European director of video for AOL, which is a customer.
When people come across video ads from Realeyes clients online, they are prompted to allow access to the camera. About 2% of people asked agree to participate, Jäätma said. Once the person agrees, the camera turns on, starts recording and uploads the video to space on Amazon Web Servers, and the Realeyes algorithms get to work.
The technology can automatically detect a person’s gender and age bracket, but Realeyes could merge the information with other data sets to build a more specific profile of a person.
The company and its partners are admired in the field, said Rosalind Picard, an MIT professor who is a leading researcher in the field.
The work raises privacy concerns however about how the information could be used, particularly in Europe, where privacy laws are more stringent. Jäätma said the company takes privacy seriously, undergoes audits and gets explicit consent from people they are filming.
Privacy advocates say that if people who agree to give access to their webcam truly understand what they’re agreeing to, there shouldn’t be a problem. But that often isn’t the case, said Anna Fielder, board chair of Privacy International, a London-based advocacy group.
“If the company said to people, ‘We are mapping your face in order to see how you react to advertising, do you agree to be mapped in that way?’ and they said yes I’d love it, we’d have no objection,” she said.
Realeyes tells people that allowing access means the company will be able to see how they respond emotionally during the video to “help us make better content in the future.” There’s also a link to a full privacy policy of more than 1,000 words.
That policy says all personal information – defined as “anything that can be used to identify you, including facial recordings and IP addresses” – are kept for as long as required. Jäätma said that practically that means about five years.
“The database itself is becoming an interesting asset,” he said. “We can go back to those huge data sets that we now have – it’s really massive by now – to use that to train next generation of algorithms. For that purpose we actually keep all those recordings in our data centers.”
Jäätma said he sees the company moving into other areas. In particular, he said he would like to build a mental-health product that helps people become happy, and stay happy.
Imperial College professor Maja Pantic, who is leading the research, said she hopes that one day the technology will help autistic children communicate or help people who are depressed.
No comments:
Post a Comment