The Delhi Police is set to significantly expand its use of facial recognition — from localised experiments to full-scale, centralised operations — as part of a major shift towards AI-powered surveillance.
Police have major ambitions from this citywide surveillance overhaul that will kick in from June: not just facial recognition, but number plate identification and predictive analytics too. But the move has also raised concerns about a legal framework for the deployment of the technology and potential misidentification of suspects.
Currently, the North and North-West Delhi police districts run vans equipped with an Israeli software which is capable of scanning faces on streets and flagging suspects. Now, police are setting up a state-of-the-art Integrated Command, Control, Communication & Computer Centre, or C4I, that aims to unify and exponentially enhance surveillance capabilities across the city using facial recognition. The software has been trained and developed by a research organisation under the Ministry of Electronics and Information Technology.
While Apar Gupta from the Internet Freedom Foundation termed it “a quantum leap in the state’s ability to identify and monitor individuals”, he pointed out that by combining facial recognition with other data streams, authorities can build detailed profiles of individuals. “Facial recognition is uniquely intrusive: real-time, automated identification at scale, erasing public anonymity,” he said.
The Israeli technology was originally acquired by the Delhi Police in 2018 to match pictures of lost and found children. According to an RTI filed by the Internet Freedom Foundation, the Delhi Police used the software to investigate cases such as the 2020 North-East Delhi riots and the 2022 Jahangirpuri violence. Facial recognition cameras have also been used in high-security events such as Republic Day and Independence Day parades since 2018, along with the G20 summit in 2023.
Now, the C4I, developed by the Centre for Development of Advanced Computing, will serve as Delhi Police’s nerve center for AI-driven crime detection. It will receive live feeds from 10,000 high-resolution CCTV cameras installed across the city and integrate footage from “legacy networks” maintained by municipal bodies, RWAs and the PWD department.
These feeds will be analysed in real time, with AI models capable of identifying over 20 faces in a crowd, even under partial visibility or disguised appearances.
It will be equipped with features such as gunshot detectors, estimating the number of people in a crowd, identifying people who have collapsed in a certain place and alerting police when someone seems to be in trouble.
The C4I will include a data centre and two emergency operation centres — which will be equipped to flag crimes in real time to district headquarters and local police stations — and will be able to monitor up to 1,000 live CCTV streams simultaneously.
From facial recognition and automatic number plate recognition to predictive behavior analytics, the system is being designed to handle it all. “We are aiming for match times under five seconds,” said B S Jaiswal, who was Joint Commissioner (Tech and PI). Jaiswal was recently transferred to his next assignment as Joint CP (Central Range).
To support the vast volume of video intelligence, the Picture Intelligence Unit (PIU) will maintain detailed audit logs and access data from a host of national databases, including e-Challan and even telecom and banking records. The PIU will also be tasked with tagging photos from raids, newspapers and public submissions to continuously improve the AI’s recognition accuracy.
Despite the ambition, Jaiswal admits that there are challenges: from poor camera angles to weather interference, and from demographic bias in AI models to privacy-compliant training datasets. But he says: “Like beat cops learn to spot anomalous situations, our machines will learn too.”
