Why India’s CCTV cameras need AI to actually fight crime

This article is authored by Atul Rai, CEO & co‑founder, Staqu Technologies.

Walk through any Indian city today and it feels like every corner is under watch from traffic junctions and metro stations to markets, offices, schools, and housing societies. With over 84,000 CCTV cameras already installed across 100 Smart Cities and integrated into citywide command-and-control centres, India has built one of the most expansive urban surveillance infrastructures in the world. Yet, every time an incident occurs, the familiar line resurfaces: “CCTV footage is being examined,” and by the time hours or days of video are manually reviewed, the trail often goes cold and the perpetrator is untraceable.

AI (iStock)

This paradox exposes a hard truth: the problem is not the lack of cameras, but the way they are used. Conventional cameras, no matter how many are deployed, are essentially dumb devices until a human steps in to watch, interpret, and act on the footage. They record everything but understand nothing. In effect, cities have built vast archives of visual data that are incredibly rich in evidence yet painfully slow and costly to exploit when time is of the essence.

India’s surveillance build-out has been accelerated by programmes like the Smart Cities Mission and by a rapidly expanding CCTV and electronic security market. Government data indicates that more than 84,000 cameras are already linked to Integrated Command and Control Centres (ICCCs) across 100 Smart Cities for public safety, traffic enforcement and incident monitoring. Industry analyses show India’s CCTV and video surveillance market running into several billion dollars and growing strongly, signalling that investments in visual infrastructure will only deepen in the coming years.

Yet, the everyday policing experience often tells a different story. When a theft, assault, or hit-and-run occurs, investigators typically piece together footage from multiple public and private cameras, then manually scrub through hours of recordings to identify one critical moment. In many cities, a significant share of footage requests cannot be serviced on time due to fragmented ownership, storage limitations, and manpower constraints, leading to low resolution rates in categories like snatching and street crime despite the presence of cameras.

The core limitation is architectural: traditional CCTV systems were designed as recording tools, not as real-time decision systems. They capture and store data but rely on human eyes for detection, correlation, and judgment. In practical terms:

• A camera can “see” a person brandishing a weapon, but it cannot raise an alarm on its own.
• It can record a snatching incident, but cannot automatically flag the suspect’s face or clothing for tracking across other feeds.

This human-in-the-loop model is not scalable in cities that generate millions of hours of video each day. Even with ICCCs and monitoring rooms, a finite number of operators cannot meaningfully track hundreds or thousands of live feeds simultaneously. As a result, cameras become “silent witnesses”, useful after the fact, often in a limited way, but rarely preventative in real time.

The recent case on the Purvanchal Expressway in Uttar Pradesh starkly illustrates the privacy risks of this human-dependent model, where unchecked access turns surveillance into a weapon against citizens. A toll plaza assistant manager allegedly exploited high‑resolution CCTV feeds by zooming into parked vehicles to secretly record commuters’ most private moments—including those of a newly married couple—then used the footage to demand extortion payments and even leaked it online after one such shakedown. This incident, which sparked complaints all the way to top authorities, reveals how entrusting cameras to human operators without oversight exposes everyday people to harassment, blackmail, and dignity violations far worse than any oversight gap in crime detection.​

Artificial Intelligence (AI) changes this equation by adding a real-time analytical brain to the existing visual infrastructure. In an AI-enabled setup platform can analyse hundreds of thousands of frames per second with sub-second latency, turning raw footage into searchable, actionable intelligence. Investigators can perform text-based video searches such as “man in red T‑shirt” or “person with green bag” or run facial recognition and object detection across distributed camera networks within minutes, instead of manually scrubbing through hours of video.

Crucially, AI also enables policy‑driven controls that mitigate the privacy invasions enabled by unchecked human viewing, as seen on the Purvanchal Expressway. Features like role‑based access restrictions, automated audit logs of all footage views and exports, automated masking of sensitive areas or faces in non-threat scenarios, and real‑time alerts for anomalous zooming or prolonged scrutiny can enforce privacy-by-design without relying on individual ethics. In this framework, the system itself detects and blocks potential misuse by operators, safeguarding citizens from internal threats while empowering proactive crime prevention

Across India, AI-enabled CCTV is already demonstrating tangible law-enforcement benefits. AI video analytics and facial recognition are being used by police forces to identify suspects in large robbery cases, trace getaway routes, and reconstruct crime timelines in a fraction of the time earlier required. During high-security events, AI systems monitor thousands of live feeds to detect suspicious behaviour, unattended objects, crowding, or perimeter breaches and generate real-time alerts for security teams. In some cities and critical facilities, AI-powered cameras now automatically flag loitering, vandalism, or the presence of weapons, enabling proactive intervention rather than delayed reaction.

The lesson for India’s surveillance strategy is clear: Adding more lenses without adding intelligence will only deepen the archive problem. The country is already on a strong growth trajectory for AI-enabled CCTV, with the India AI CCTV market valued at around $ 827 million in 2023 and projected to grow more than four-fold to about $ 3.67 billion by 2030. This reflects a strategic shift from simply installing cameras to deploying systems that can detect, correlate, and alert in real time.

For policymakers, city administrators, and enterprises, the imperative now is to treat AI not as an add-on but as the core of any new surveillance investment. Cameras must evolve from passive recorders into active collaborators in safety systems that can identify threats, support officers on the ground, and help prevent heinous incidents before they escalate. Otherwise, the risk is clear: India will continue to spend heavily on technology that watches everything, yet meaningfully protects too little. Smart cities demand smart cameras eyes that do not just see, but also understand and help solve crime.

This article is authored by Atul Rai, CEO & co‑founder, Staqu Technologies.