An Indian startup supplying facial recognition for a new airport-clearance service says its algorithm is right 99 percent of the time. The company, Dataevolve, wrote the code for Digi Yatra, which is a contactless biometric security service still taking off in India. It also has suffered serious growing pains.
The company trained its algorithm on a 600,000-image dataset made up of Indian nationals, according to business publication Moneycontrol. Building the dataset with images from the same demographics as Digi Yatra’s users avoids the accuracy differentials that have affected some face biometrics algorithms.
There are three or four thousand travellers using Digi Yatra on an average day at four airports. The app, on Android and iOS, sends a selfie and a person’s Aadhaar image to be verified on Amazon’s Web Services. The company maintains that verification is impractical for a phone.
Credentials, saved onto wallets, are issued from Aries Hyperledger, which reportedly is optimised for this kind of work. The Digi Yatra Foundation has tried to assuage travellers’ fears of a data breach, saying biometric data is stored locally, making it less likely to be the subject of a massive breach.
