Video Analytics in India: How AI CCTV Analytics Works, Use Cases and Cost (2026)

Video analytics is software that watches CCTV footage for you, detecting people, vehicles, objects and events in real time and raising an alert or report when a rule is met. In India it now runs city police networks, railway stations and retail stores, and it can usually be added to existing IP cameras.

Key takeaways

• Video analytics turns passive CCTV into a system that detects and alerts, instead of footage someone checks only after an incident.

• AI (deep learning) analytics recognise people, vehicles, faces, number plates, crowds and behaviour; older rule-based analytics mostly detect motion.

• Analytics can run on the camera (edge), on a local server, or in the cloud; most Indian enterprises end up with a hybrid.

• Cost depends on the number of camera streams, the analytics you need, processing hardware and integration, not just the camera.

• Under the DPDP Act, 2023 and the DPDP Rules, 2025, video that identifies people is personal data, so plan purpose, notice, access and retention from day one.

In October 2026, Gujarat Police hosted what it called India's largest AI-CCTV hackathon. More than 3,100 teams registered and about 1,000 solutions were submitted, with teams demonstrating number plate recognition, vehicle tracking and watchlist matching on live CCTV feeds (DD India, 2026). The message is clear: India has installed lakhs of cameras, and the next step is making them understand what they see. That is what video analytics does.

This guide explains what video analytics is, how it works, the main types, real Indian use cases, edge vs cloud options, privacy rules and what drives cost.

What is video analytics?

Video analytics is software that automatically analyses live or recorded video to detect, classify and track objects and events, then triggers alerts, reports or searches. Instead of a guard watching 40 screens, the software watches every feed and only asks a human to look when something matters.

The output is not just an alarm. Good video analytics also creates metadata, such as "red car, 10:42, gate 2", so you can search hours of footage in seconds.

Basic (rule-based) vs AI video analytics

What is intelligent video analytics (IVA)?

Intelligent video analytics is the industry term for AI-based video analytics. It classifies objects, follows them across frames and understands context, for example a person who stays near an ATM for 15 minutes, rather than just "something moved". Most modern AI CCTV camera and software platforms in India now sell IVA rather than basic motion detection.

How video analytics works

Video analytics works in five steps: capture, decode, detect, apply rules and act. AI based video analytics follows the same flow; only the detection step is smarter.

1. Capture: The camera streams video, usually over RTSP or ONVIF, to the analytics engine. This can be inside the camera, on an NVR or server, or in the cloud.

2. Decode: The engine decodes the H.264 or H.265 stream into frames. Many systems analyse a few frames per second rather than every frame to save processing power.

3. Detect and classify: AI models find objects in each frame (person, car, bike, bag, face, number plate) and track them as they move.

4. Apply rules: You set what matters, such as a virtual fence on a boundary wall, a queue zone at a billing counter, or a list of blacklisted number plates.

5. Act: When a rule is met, the system sends an alert to a control room, mobile app or VMS, starts high-quality recording, updates a dashboard, or pushes data to another system such as an ICCC, access control or ERP.

A human should verify important alerts before acting. Even Delhi Police's Safe City system routes AI alerts to operators at its C4i command centre, who check them before patrol units are sent (ThePrint, 2025). An AI video analytics platform should make that verification step fast, with a short clip attached to each alert.

Types of video analytics

The most used video analytics in India are intrusion detection, ANPR, facial recognition, crowd and people counting, and safety analytics such as PPE detection. Each one needs the right camera setup to work well.

For deeper guides on two retail analytics, see our posts on queue monitoring and footfall counting.

Video analytics use cases in India

Indian police, railways, metros, retailers and factories already use video analytics for safety, traffic and operations. Here is how it shows up by sector.

Smart cities and police

Delhi's Safe City project switched on its first phase of more than 3,500 AI-enabled cameras, drawn from a pool that includes 1,622 ANPR cameras and 370 facial recognition cameras, alongside 200 gunshot detection systems (ThePrint, 2025). In Raipur, traffic police are using the Trinetra traffic management control room with cameras and drones this festive season, with automatic congestion alerts sent to field teams (The Hitavada, 2026). Read more on AI surveillance for Indian police and how a smart city surveillance system is designed, or see our government work.

Crowd management at large events

For the Nashik Simhastha Kumbh 2027, where more than 12 crore pilgrims are expected (Free Press Journal, 2026), planners are building AI crowd monitoring across CCTV, drones and sensors. It will alert officials when crowd density reaches critical levels, so they can divert pilgrims or send emergency teams (Free Press Journal, 2026).

Transportation: railways and metro

AI-enabled video surveillance, including intrusion and loitering detection and facial recognition, has been installed at 1,874 railway stations in 2025-26, according to an official statement (Ommcom News, 2026). Kolkata Metro's Blue Line project uses 1,678 cameras with alerts for unattended baggage, perimeter breaches and unusual crowd behaviour, plus crowd heat maps and platform dwell-time analytics (Free Press Journal, 2025). See our transportation solutions.

Other industries

Edge vs server vs cloud video analytics

Edge analytics runs inside the camera, server analytics runs on a local GPU server or NVR, and cloud analytics runs in a data centre; each trades cost, bandwidth and flexibility.

Edge AI is growing in India. In December 2025, CP Plus and Qualcomm announced an AI video security platform that processes video on-premise using edge boxes in the VMS, with commercial availability expected in early 2026 (DQ Channels, 2025). Whatever you choose, the network matters as much as the software, which is why video analytics projects should be planned with your IT infrastructure.

Can you add video analytics to existing CCTV?

Usually yes: most IP cameras that stream over RTSP or ONVIF can feed a CCTV video analytics server without being replaced. Check these points first:

1. Camera type: IP cameras work directly. Analog (HD-CVI, AHD) cameras need a DVR or encoder that can stream, or an upgrade.

2. Resolution: 2 MP is enough for intrusion and counting; ANPR and facial recognition usually need higher resolution and a dedicated angle.

3. Placement and lighting: The right angle matters more than megapixels. Backlit doors and dark car parks cause most misses.

4. NVR or VMS compatibility: Confirm the analytics can pull streams from your recorder or cameras.

5. Network: Each analysed stream uses bandwidth to the processing point.

6. Camera uptime: Analytics fail silently if a camera is offline or covered. CCTV health monitoring keeps the cameras that feed analytics working.

7. Storage: Event clips and higher-quality recording on alerts need planned CCTV storage.

Accuracy, false alarms and limits

Video analytics accuracy depends more on camera placement, lighting and tuning than on the software brand. Be wary of any vendor that quotes a single accuracy number without testing on your cameras.

• False alarms: Shadows, headlights, rain, insects on the lens and stray animals still trigger alerts if zones and sensitivity are not tuned.

• Night and weather: IR range, glare and fog reduce detection. ANPR in particular needs IR and the right shutter speed.

• Facial recognition limits: Accuracy drops sharply with side angles, masks, low light and low resolution.

• Human in the loop: Alerts should go to a trained operator who verifies them before action, as Delhi Police does at its command centre.

• Pilot first: Run 2-4 weeks on real sites, count true and false alerts, then tune before scaling.

Video analytics, privacy and the DPDP Act

If your video analytics identifies people, the footage and outputs are personal data under India's Digital Personal Data Protection Act, 2023. The DPDP Rules were notified on 14 November 2025 with an 18-month phased compliance timeline (The Tribune, 2025), so organisations deploying analytics now should design for them.

• Purpose: Write down why each analytic is used (security, safety, operations) and do not reuse data for unrelated purposes.

• Notice: Use clear signage where cameras and analytics operate.

• Minimise: Use counting and heat maps without identifying people wherever identity is not needed.

• Access and retention: Limit who can see faces and plates, log access, and delete data on a fixed schedule.

• Vendors: Put data protection terms in contracts with analytics and cloud providers.

Public scrutiny is rising. In October 2026, the Internet Freedom Foundation questioned the transparency of a Delhi Police tender for facial recognition cameras around protest areas (Biometric Update, 2026). Private deployments should expect similar questions from employees and customers. This section is general information, not legal advice; our AI consultation team can help you plan a privacy-aware rollout with your legal advisers.

Video analytics cost in India: what drives the price

Video analytics cost in India depends on the number of camera streams, which analytics you run, the processing hardware and integration work. Two sites with the same camera count can have very different budgets.

Public projects show the scale of full systems. Delhi Police's July 2026 tender for 338 facial-recognition-enabled cameras and 47 number plate readers in two districts has a budget of ₹15.66 crore (Biometric Update, 2026). Kolkata Metro's Blue Line AI-CCTV project, with 1,678 cameras and 240 TB of central storage, costs ₹25.34 crore (Free Press Journal, 2025). These figures include cameras, network, storage and installation, so treat them as scale references, not analytics software prices. For a business site, the honest answer is a site survey: get a site-specific estimate.

How to choose video analytics software

Start from the problem you need solved, then test video analytics software on your own cameras before buying. When comparing video analytics companies in India, use this checklist:

1. Define use cases and KPIs: for example, "cut false perimeter alarms" or "measure store conversion".

2. Audit your cameras: list models, resolution, angles and which ones can be reused.

3. Check compatibility: ONVIF/RTSP support, your NVR or VMS, and APIs for other systems.

4. Choose the architecture: edge, on-premise server, cloud or hybrid, based on sites and bandwidth.

5. Run a pilot: 2-4 weeks on real sites, measuring true and false alerts.

6. Check privacy controls: role-based access, masking, audit logs and retention settings.

7. Check local support: on-site service in your cities, tuning and AMC terms.

How Transline helps

Transline Technologies designs, deploys and supports video analytics for enterprises and government across India, usually on top of the cameras you already have.

• CamPulse: our AI video analytics platform that turns existing CCTV into real-time security, safety and operations alerts.

• StorePulse: retail analytics for footfall and in-store insights.

• CheckCam: CCTV health monitoring so the cameras feeding analytics stay online.

• CamStore: storage for recordings and alert clips.

• Surveillance solutions and AI consultation to plan cameras, network, analytics and privacy together.

Want to see what your cameras could detect?Talk to our team for a site assessment and a pilot plan.