Of the 100 command centres built, 30 work on traffic and transport today. Central operations funding for these assets ended after a June 2025 advisory, and cities are now working out what comes next on their own.
The cameras are bought. The fibre is laid. The control rooms are built. The open question is what happens after a camera sees something.
A systematic review of 161 CCTV evaluations, 76 with usable data, found an average crime reduction of about 13 percent. The average hides the finding that matters.
Actively monitored cameras, with a dispatch pathway, cut crime by about 15 percent. Passively recorded cameras, stored but not watched, showed no measurable effect. Narrow deployments, like car parks, performed best, at around 37 percent. The same technology produced opposite results in the UK and the US: a 21 percent reduction in one, none in the other.
Effect sizes do not travel well across contexts. What does travel is the mechanism behind the ones that worked: a legal instrument, a named responder, and a way to close the loop.
A few things recur across the deployments that produced results:
Camera evidence needs to hold up as evidence by itself, without an officer having witnessed the violation in person.
The road, the policy, the enforcement, and the payment need to sit with a single owner rather than split across departments.
An alert only does something if a specific person owns what happens next. A live feed on a screen, on its own, is not a response.
The camera needs an existing database to match against. Barcelona's enforcement ran against Spain's national vehicle registry rather than a new one built for the purpose. India has VAHAN for this.
A good example is London's low-emission zone, which paired the charge with scrappage support for vehicle owners affected by it. Compliance reached 96.7 percent.
New York's camera law ties renewal to published outcome data. It has produced one of the more complete public datasets on the subject.
Each of these is a decision about how a program is run, not a line item in a procurement spec, and each costs less than the cameras themselves.
A camera increases how many notices get issued. Whether that changes behaviour depends on follow-through: whether the notices are actually collected on.
In India, national e-challan realisation was 77 percent in 2019. By 2024, detection capacity had scaled roughly sevenfold, and realisation had fallen to 24 percent. States running the same cameras under the same law see very different recovery rates, from over 70 percent in some to under 15 percent in others. The difference comes down to recovery.
Camera networks fail in ordinary, avoidable ways. One state's traffic sensors ran at 28 percent uptime against a 97 percent service-level agreement. In another city, 22 of 710 cameras were working at a given time. One network, built at a cost of ₹232 crore, went offline over an unpaid telecom bill of ₹5 lakh.
Uptime is rarely written into Indian camera contracts as a payment condition. It should be, alongside a published baseline: how many cameras are actually working, what footage retention looks like by location, what the realisation rate is by offence type. Establishing that baseline takes about four weeks and no capital spend, and it gives everything that follows something to be measured against.
India has no dedicated statute governing public CCTV or ANPR surveillance, though ANPR is recognised for specific purposes under the country's road-traffic and tolling rules. The Supreme Court's Puttaswamy judgment set the test that applies instead: any government intrusion into privacy needs to be legal, serve a legitimate purpose, and be proportionate.
India's Digital Personal Data Protection Act will add a fuller data-protection layer once its substantive provisions take effect in May 2027, with meaningful, though not blanket, exemptions for state and law-enforcement processing.
Regulation specific to camera networks will likely follow, as it usually does, once deployments are widespread enough to force the question. Until then, the proportionality test is decided less by statute and more by how a system is designed and what a contract permits. Cities specifying camera networks now are making that call today, whether or not they intend to.
We recommend starting with one corridor, not a city-wide platform: measuring what already exists and publishing the number, then instrumenting one stretch with a known problem for 90 days, with uptime and accuracy written in as contractual commitments. From there, replication should follow the evidence.
The technology is rarely the hard part. The baseline, the responder, the recovery chain, and the contract that holds them together are.
