New Delhi: Making a giant leap in the modernisation of the police force, Delhi Police is installing state-of-the-art facial recognition systems (FRS) to fight crime throughout the Capital. Part of the overarching smart surveillance strategy, this program has already been yielding tangible outcomes in places such as North and Northwest Delhi, where the system has been rolled out aggressively since 2024.
The revolution started with mobile FRS vans patrolling high-crime zones. Equipped with a range of high-resolution cameras, including specially modified Automatic Number Plate Reader (ANPR) units, the vans are now able to recognize people through facial scans. The system analyses video in real time, alerting on faces that match records in the police’s criminal database with more than 60 per cent similarity. Alerts are sent immediately to on-duty officers.
This live identification has already been instrumental in many arrests. In one instance, a robbery suspect was caught only three days after a holdup in Fateh Puri, based on a match through facial information derived from the data recovered from the CCTV.
Inside the van, officers watch live feeds on dual screens. The software doesn’t simply match facial features; it estimates age and height and even detects faces partially hidden by masks or disguised by facial hair. The system’s ability to identify people despite these obstructions has made it a valuable tool in shrinking suspect lists rapidly and efficiently.
Nonetheless, FRS has its limitations. Real-world scenarios such as dim lighting, off-angle camera angles, blur due to motion, and alterations of appearance due to aging or disguise can decrease accuracy. Regardless of these, police officials assert that ongoing machine learning updates will make the system increasingly reliable in the long run.
The Delhi Police, in partnership with the Centre for Development of Advanced Computing (C-DAC), is now expanding this technology throughout the city. This rollout comprises the deployment of 10,000 AI-based cameras under the “Safe City” project. These streams will be routed to a centralized Integrated Command and Control Center (C4I), which will monitor all district-level video analytics.
This command center, capable of processing feeds from 1,000 cameras simultaneously, will also be connected to legacy networks operated by municipal councils, market associations and
public transport agencies.
Apart from fixed cameras, 88 mobile units known as Prakhar Vans will also keep scanning spaces for real-time threat detection.
Adding to this system is the Picture Intelligence Unit (PIU) that classifies pictures by having unique identifiers such as age, gender, tattoos, or scars. It also accesses several databases of the government to develop an extensive searchable index of suspects.
To address the technical hurdle of training AI on Indian populations, the FRS models are fine-tuned on local datasets. This guarantees enhanced recognition accuracy of Indian faces and corrects high error rates associated with foreign-trained models.
Striving for the perfect trade-off between false acceptances and rejections, the Delhi Police is seeking a lower Equal Error Rate (EER), indicating peak accuracy. With ongoing advances in AI and data processing, facial recognition has the potential to become a bedrock in Delhi’s crime control and public safety initiatives.