Delhi Police film protests, run its images through face recognition software to screen crowd

Before December 22, AFRS was used only thrice — twice at the Independence Day parade and once at the Republic Day parade.

Sources said Delhi Police has so far created a photo dataset of over 1.5 lakh “history-sheeters” for routine crime investigation. Another dataset meant for monitoring sensitive public events has over 2000 images of terror suspects and — the latest addition — alleged “rabble-rousers and miscreants”.

ALSO READ | Drones to bandobast to peace meetings: UP has a quiet Friday

The use of AFRS by police assumes significance given ongoing protests across the city against the new citizenship law and the National Register of Citizens — events where police were seen videotaping proceedings.

For long, Delhi Police has been filming major protest events in the city. This footage is now being fed to AFRS which extracts “identifiable faces” of the protesters to its dataset. Once extracted, say sources, the lot is manually screened to identify and retain “habitual protesters” and “rowdy elements”.

This dataset of “select protesters”, said sources, was put to use for the first time to keep “miscreants who could raise slogans or banners” out of the Prime Minister’s rally last Sunday.

“Each attendee at the rally was caught on camera at the metal detector gate and live feed from there was matched with the facial dataset within five seconds at the control room set up at the venue,” explained a police official involved with the security arrangements for the PM rally.

Delhi Police drones hover over a anti CAB protest at ythe Jorbagh in New Delhi on friday. Express Photo by Tashi Tobgyal New Delhi 271219

Contacted by The Indian Express, a Delhi Police spokesperson said in an email: “Any FRS (facial recogniton software) would hold its utility only if the related database was qualitatively and quantitatively substantial. The initial focus was on security related aspects of the Independence Day and Republic Day arrangements, achieved by building up-to-date datasets of terror suspects etc.

“In the next stage, we focused on law and order also and accordingly expanded the datasets to those with known criminal records of relevant categories and also to law and order suspects, identified through extensive archival videography and behaviour analysis at sensitive public protest venues. We have used these datasets last Sunday based on credible intelligence inputs about possible disruptions.”

Asked what checks and balances were built in to avert misuse of images and satisfy privacy concerns, the spokesperson said: “Without getting into operational details, Delhi Police assures that best industry standard checks and balances against any potential misuse of data are in place. Such datasets are not perpetual and are revised periodically. Racial or religious profiling is never a relevant parameter while building these datasets.”

At the present capacity, the AFRS version that Delhi Police have can handle different facial datasets adding up to 3 lakh faces at a time. “This can be enhanced to handle 9 lakh units,” said one of the promoters of Delhi-based Innefu Lab that supplied the system to the police.