Loading…
Loading…
Smart city video analytics in India: what city control rooms need from AI, from traffic and crowds to privacy and procurement, and why local platforms matter.
Indian cities have spent a decade installing cameras. The next decade is about making those cameras useful. AI video analytics turns hours of footage nobody watches into alerts, evidence and decisions, in real time.
For governments, police and city operators, this is no longer a future concept. It is an active and fast-growing market, and the choices made now will shape urban safety for years.
The growth is structural, not speculative. Government programmes such as the Smart Cities Mission, dense urban populations and steady camera rollouts across large and mid-sized cities all push in the same direction.
Cameras alone do not create safety. The value lies in the analytics layer on top: software that watches every feed, understands what it sees and acts on it. That is where investment is heading next.
Indian cities operate in conditions very different from Singapore or London. Extreme density, uneven infrastructure and many languages create demands that imported systems rarely meet out of the box.
In practice, city operators converge on three needs. They want reliable real-time analytics across very large camera estates. They want it to work with the cameras and recorders they already own. And they want alerts that lead to action, not more screens to watch.
City projects cover thousands of cameras across wide areas, on budgets that rarely match those of Western cities. Doing more with less is the starting point, not a constraint.
Solutions that cannot deliver enterprise performance at sensible cost do not survive procurement. That reality shapes every design decision, including where video lives and how it is processed.
Data sovereignty matters for public safety footage. Many city and police buyers will not send sensitive video through external cloud servers.
Visylix, our automated AI video surveillance platform, runs entirely on the customer's own hardware: on premise, at the edge, in a hybrid setup or fully air-gapped. It ships as a Docker image, and video never leaves the building. Its AI and its Radha copilot run on site too.
Alerts mean little if the people receiving them cannot read them quickly. A single control room may include staff who work best in Hindi, Tamil, Telugu, Kannada or Bengali.
That is why the Visylix interface is available in 55 languages, including 13 Indian languages. Radha, its AI copilot, answers in English, Hindi and Hinglish, so operators can ask questions the way they speak.
City cameras face monsoon rain, dust, fierce summer heat and dense winter fog. All of these degrade image quality and challenge models trained on clean footage.
Strong platforms are built for this. Visylix learns each camera's normal, so its motion, crowd and anomaly alerts adapt to the scene. When an operator marks a false alarm, thresholds tighten automatically, never below the limit the operator has set. Small, distant subjects are still detected.
City surveillance has to fit national frameworks, from MeitY guidelines and data localisation norms to the Smart Cities Mission's expectations for Integrated Command and Control Centres (ICCCs).
A platform that cannot pass its events to the city's other systems is not ready for city-wide use. Visylix sends alerts through signed webhooks, MQTT and connections to security information systems (SIEM) and control room software (PSIM), and offers 500+ REST APIs for integration.
Generic feature lists fall short in Indian conditions. The capabilities that make a difference are the ones that hold up in dense, varied, fast-moving scenes.
Visylix includes 22 AI analytics built in-house, covering people, vehicles, safety and sound. Cities choose which analytics run on which cameras, and set detection zones per camera and per analytic. The sections below cover the ones city buyers ask about most.
Automatic number plate recognition (ANPR) reads vehicle plates from video. It is one of the most important tools in city surveillance, used to enforce traffic rules, find stolen vehicles and understand how traffic moves.
Visylix number plate recognition supports allow, block, watch and VIP lists, and traces exact plate journeys across cameras with travel times. Speed estimation flags speeding, wrong-way driving and U-turns, and parking analytics flag full lots and no-parking violations.
Crowds are where city surveillance faces its hardest test. Railway stations, temples, festivals and rallies draw dense gatherings, and the gap between a normal crowd and a dangerous one can close in seconds.
Visylix crowd detection measures density per square metre and alerts when a crowd is forming or a stampede risk appears. Pose estimation flags falls and running, so operators see trouble as it starts.
Safety gear monitoring is a growing use in industrial zones, construction corridors and city infrastructure projects. Cameras flag a missing helmet or vest as it happens, before an accident occurs.
Visylix PPE detection checks helmets and vests by default and up to nine items in total. It scores mAP@50 0.865, a standard accuracy measure, on a frozen evaluation set of 10,315 images.
Power substations, transit depots, water works and data centres often sit right next to busy neighbourhoods. Their perimeters need instant alerts, not footage reviewed the next morning.
AI intrusion detection and line crossing raise an alert the moment someone enters a restricted zone. The hard part is keeping alarm volume manageable. In Visylix, Sudarshan rules combine conditions, such as an intrusion after hours with no matching badge swipe, so operators see real threats rather than routine movement. A rule can also trigger a talkdown warning through the camera's own speaker.
A platform built in India is designed from the start for Indian scenes: diverse crowds, multilingual signage, mixed traffic and monsoon visibility. That local design is not cosmetic. It shapes how well the system performs on the street.
Imported enterprise platforms typically charge per camera or per channel, which makes city-scale estates expensive. Visylix takes a different approach.
Its Starter (₹4,999 per month) and Pro (₹9,999 per month) plans include unlimited cameras on the VMS licence. AI analytics are a separate subscription, billed per camera per month, only on the cameras a city chooses. This lets a city record and view everything, and pay for AI only where it adds value.
India's diversity is operationally significant, not only demographic. Clothing, signage, vehicle mix and crowd behaviour vary from region to region. Systems shaped only by Western footage often struggle in Indian streets, and platforms built here are designed with that variety in mind.
Indian public procurement has its own rhythm: GeM portal listings, detailed tender documents, lowest-bid (L1) evaluation and layers of compliance. Vendors who understand this process from the inside reduce friction at every stage, from technical bid to acceptance.
Remote troubleshooting alone does not work when a camera network fails during a major festival or event. Cities need partners who can respond in their time zone, in their language, with people who understand the site. This expectation increasingly shapes how vendors are judged.
The market is crowded. Global hardware makers, domestic system integrators and AI-first software platforms all compete for the same tenders.
What separates lasting winners from one-project vendors is rarely the specification sheet. It is local readiness, depth of capability and the ability to support a city long after go-live. As scrutiny of AI surveillance grows, platforms that pair performance with transparency, such as audit trails and evidence that can be independently checked, will lead.
Indian city surveillance is reaching a turning point. Connectivity is no longer the differentiator. Automation is: the systems that will deliver lasting value are the ones that watch, decide and act without waiting for someone to notice a screen.
At Aptibit, we built Visylix around that idea. It watches every feed with its AI analytics. It decides with Sudarshan rules, which combine conditions, sequences and access control events. It acts through alerts, incidents in the operator console, integrations and the Radha copilot.
Radha runs on an on-premise language model and has 111 tools for search, investigation and control. Its 13 high-impact actions always ask for confirmation, and every call is recorded in the audit log.
Automation is already reshaping how Indian cities protect people and infrastructure. The shift is from watching screens to acting on alerts, and from reviewing footage after an incident to responding while it unfolds.
The question is no longer whether AI will define city surveillance in India. It is which cities move first, and how well they do it.
Mostly by adding intelligence to the cameras cities already own, rather than replacing them. Analytics software reads the existing feeds, either in a central control room or at the edge, close to the cameras. Processing locally saves network capacity and keeps footage in the city. It also keeps costs manageable at city scale.
Its main value is speed and reach. AI can watch every feed at once and raise an alert in seconds, which no team of operators can do by eye. It also gives traffic police consistent, timestamped evidence. Results are not uniform: they depend on camera quality, coverage, clear response procedures and operator training.
The safety gains come with a real trade-off. The Digital Personal Data Protection Act sets rules on consent and purpose, but biometric collection in public spaces still lacks dedicated guidance. Cities should limit retention, restrict who can search footage, log every search and redact faces and plates when footage is shared.
Under the Smart Cities Mission, many cities have set up Integrated Command and Control Centres, the control rooms that bring traffic, safety and civic data together. Hyderabad, Surat and Delhi are often cited for large city camera networks. Traffic violation detection and congestion management are the most common uses. Results depend on execution, not ambition.
India's legal framework for AI surveillance is still forming. The Digital Personal Data Protection Act, 2023 sets the foundation for consent and purpose. There is not yet a single statute that governs facial recognition or behaviour analytics in public spaces, so cities rely on policy, procurement terms and good practice.
A typical system combines cameras, a video management system that records and displays them, and AI analytics that read the video. Common analytics include number plate recognition, face recognition against watchlists, crowd detection, intrusion alerts and traffic rule detection. Rules then decide what happens next: alert an operator, notify field teams or push events to other city systems.
Uneven infrastructure, budget limits and privacy concerns are the biggest barriers. Public trust has to be earned through transparent, accountable use. And platforms must work in real Indian conditions: dense crowds, heavy traffic, monsoon rain, dust and many languages in the same control room.