Video Management and Video Analytics: The Complete Guide

Walk into any Integrated Command and Control Centre (ICCC) in India and most of what you see on the video wall comes from cameras. Thousands of them watch junctions, markets, highways, railway stations and public gatherings. But cameras alone do not make a city safer. The value comes from two layers of software: the Video Management System (VMS), which collects, records and serves the video, and video analytics, which watches that video automatically and raises an alert when something needs a human’s attention.

This guide explains both layers in plain language, with separate sections on traffic analytics and law and order analytics, the latest advances in AI, and the role of Local Processing Units (LPUs) at the edge. Throughout, we focus on Indian stories. It is a companion to our Control Rooms: The Complete Guide, and it links to every related article we have published.

Infographic: how video analytics works, from a junction camera view with AI detections, through the Local Processing Unit, network, VMS and AI analytics to the ICCC and response, plus traffic and law and order analytics — How video analytics works: from camera to action (click to enlarge).

VMS and video analytics: what is the difference?

A VMS is the backbone. It connects to cameras, shows live video to operators, records it, stores it for a set period and lets investigators search and export footage. Video analytics is the intelligence on top. It uses computer vision, today mostly deep learning, to recognise vehicles, number plates, people, crowds and unusual behaviour, and turns them into events such as “red-light violation at junction 12” or “crowd density above threshold at gate 3”.

In a city, both feed the ICCC. The VMS puts the right camera on the video wall, analytics decides which camera matters right now, and the ICCC platform links the alert to a standard operating procedure, a map and a dispatch team. Our article on how video walls are transforming India’s smart city control rooms covers the display side, and the evolving role of control rooms in modern cities explains how these centres are changing.

Video Management Systems: the backbone

A modern VMS has to do five things well. First, camera integration: cities rarely buy every camera from one maker, so the VMS must work with many brands. The industry standard for this is ONVIF, which defines common interfaces for IP cameras and video systems. Second, recording and storage: deciding how long video is kept, at what quality, and how it is protected as evidence. Third, scalability, from a single building to tens of thousands of cameras across a state. Fourth, integration with analytics, access control, ICCC software and the video wall. Fifth, cyber security, because every camera is a network device that can be attacked.

Milestone Systems’ XProtect is a widely known example of the open-platform approach. Milestone describes XProtect as open-platform video management software that lets organisations choose from a wide range of cameras, sensors and IoT devices, says it integrates with over 1,000 third-party applications, and offers it in four variants from small sites to mission-critical installations. Its recording server feature brief explains that the recording server handles all live and recorded video, audio, metadata and events, using dedicated drivers as well as standard ONVIF and universal drivers for broad compatibility. Through its integration platform and AI Bridge, Milestone says third-party video analytics applications can receive video streams from XProtect and send their results back into the VMS as events and metadata.

Indian vendors also offer VMS platforms. i2V, for example, describes its enterprise VMS as an IP-based system for large, mission-critical deployments that runs on Windows and Linux and can connect with third-party NVRs. Cyber security deserves special attention here. Our article on cyber security take-aways from the Guwahati ITMS project looks at why surveillance networks need security by design.

Video analytics: from watching to understanding

No operator can watch hundreds of feeds at once. Video analytics changes the job from watching everything to responding to what matters. Early analytics relied on simple motion detection and produced many false alarms. Deep learning models trained on large image datasets can now tell a car from a two-wheeler, read a number plate at speed, count people in a dense crowd and notice a bag left behind, even in mixed and non-lane-disciplined Indian traffic. In cities, analytics falls into two broad families: traffic analytics and law and order analytics.

Traffic analytics

Traffic analytics is the most widely deployed form of video analytics in Indian cities, usually as part of an Intelligent Traffic Management System (ITMS). Its main functions are:

ANPR/ALPR: automatic number plate recognition to read plates, check them against stolen or wanted vehicle lists, and track vehicles across junctions.
Violation detection: red-light jumping (RLVD), stop-line violations, over-speeding, wrong-way driving, riding without a helmet, triple riding and seat-belt violations.
e-challan: violation evidence linked to vehicle records so that a challan can be generated and sent to the owner.
Counting and classification: vehicle volumes by class and lane, used for signal timing and planning.
Incident and congestion detection: stalled vehicles, accidents and queues spotted early so that teams can respond.

Vehant Technologies, incubated at IIT Delhi, describes OKEAN TrafficMon as an AI and machine learning suite that detects red-light jumps, over-speeding, no-helmet riding, triple riding and wrong-lane movement in real time, alongside traffic flow monitoring and ITMS functions. On its website, Vehant publishes client testimonials describing ANPR, violation detection and e-challan deployments in cities including Gurugram, Prayagraj, Ranchi and Raipur; these are Vehant’s own published accounts.

i2V lists red-light and speed violation detection, number plate recognition, lane discipline monitoring, no-helmet and triple riding detection in its Intelligent Traffic System, with an integrated enforcement backend and auto-ticketing. Intozi describes Ikshana, its AI video analytics platform, as automating detection of red-light, ANPR, speed and lane violations, with NIC VAHAN integration for e-challans. Intozi also offers an automatic traffic counter and classifier (ATCC) and congestion and incident detection that it says can run on existing CCTV feeds. According to Intozi’s published case study, its incident, speed and violation detection systems have been deployed on National Highways Authority of India projects.

For a city story, see our posts on Guwahati’s ITMS deployment and its cyber security lessons.

Law and order analytics

For police and city security teams, video analytics helps spot risks before they become incidents and speeds up investigations afterwards. Common functions include:

Crowd density and counting: estimating how many people are in an area and alerting when a threshold is crossed.
Intrusion and perimeter alerts: people or vehicles entering restricted zones.
Abandoned object detection: bags or packages left unattended in stations, markets and venues.
Face recognition: matching faces against watchlists of wanted or missing persons.
Behaviour analytics: movement against the crowd flow, barricade jumping, loitering and similar patterns.
Vehicle screening: under-vehicle scanning at entry points to sensitive buildings.

i2V lists intrusion and perimeter alerts, abandoned object detection, crowd density, face recognition with watchlist matching and camera tamper detection among its analytics. Intozi offers crowd estimation and density analytics and a face recognition system. Vehant’s OKEAN platform covers AI video intelligence, and its NuvoScan under vehicle scanning systems capture images of a vehicle’s underside at entry gates for security checks.

The biggest Indian example is the Maha Kumbh 2025 in Prayagraj. The Press Information Bureau reported that the Mela Authority’s ICCC monitored the event through more than 3,000 cameras and underwater drones, with 1,800 AI-enabled cameras tracking crowd density and sending automatic alerts. The ICCC in-charge told PTI, as reported by Hindustan Times, that alerts popped up when density crossed a marked threshold, or when there was a barricade breach or fire, and that 13 contingency schemes were defined in the SOPs. The Times of India reported that deep learning models were used for crowd density estimation and entry-exit head counts, together with person and vehicle attribute recognition.

Face recognition raises real privacy questions. India’s Digital Personal Data Protection Act, 2023 now governs digital personal data, and the government notified the DPDP Rules, 2025 on 14 November 2025, with provisions coming into force in phases. Cities and police forces using analytics will need clear purposes, retention limits and access controls. Our article on data privacy and video surveillance goes deeper, and our older post on NVIDIA’s facial recognition camera tie-up shows how long this debate has been running.

Advancements in AI-based video analytics

Video analytics is changing quickly. Four advances stand out.

Natural-language search: instead of scrolling through hours of footage, an investigator can type a question. i2V describes IRIS as an “agentic AI” layer where teams ask questions in ordinary language, get matching moments back with supporting footage, and trace each result to its camera, time and confidence level.
Vision-language models and summaries: models that understand both images and text can describe what happened in a clip. NVIDIA’s Video Search and Summarization blueprint combines computer vision with vision-language and large language models to search, summarise and answer questions about live or archived video, and to verify alerts.
Multi-camera correlation: linking detections across cameras, locations and time so that a vehicle or event can be followed across a city rather than seen on one camera at a time, as i2V describes for IRIS.
Fewer false alarms: better-trained models and alert verification steps cut down the nuisance alerts that make operators ignore the system, which in practice is what decides whether analytics is used or switched off.

These tools still need people in charge. An AI summary or search result is a lead to check, not proof, and every alert should be verified by a trained operator before action is taken.

Local Processing Units (LPUs): analytics at the edge

Sending every high-resolution video stream from every junction to a central data centre is expensive and slow. A Local Processing Unit (LPU) solves this by running analytics close to the cameras: an industrial-grade edge server or appliance, typically fitted with AI accelerators such as GPUs or NPUs (NVIDIA Jetson-class modules are a common example), housed in a junction box, a building’s equipment room or a police station.

Indian smart city tenders already specify this. Guwahati Smart City Limited’s ITMS tender technical specifications require an LPU “placed within field junction box of the camera placement”, state that “all ITMS application should be working on local processing unit”, ask the LPU to send processed data to the central software, and require seven days of local storage. NVIDIA describes its Jetson embedded systems as platforms for running AI at the edge, and vendors increasingly offer edge options: i2V describes its AI Box as edge AI computing for real-time video analytics that supports CPU, GPU and third-party accelerators, and says its ANPR supports edge and server deployment.

The benefits of LPUs are practical:

Bandwidth: only events, metadata, number plates and short evidence clips need to travel to the ICCC, not every full stream.
Latency: a violation or intrusion is detected at the junction within moments, without a round trip to the data centre.
Resilience: if the network link to the ICCC drops, the LPU keeps detecting and storing evidence locally and syncs when the link returns.
Privacy: raw video can stay on site, with only the necessary data sent onwards, which helps with data minimisation.

The trade-off is that hundreds of LPUs spread across a city need remote health monitoring, software updates, physical protection and cyber security. Our article on 5G, edge computing and video surveillance explores the network side.

Indian stories in focus

Recent announcements show how fast Indian cities are adding AI to their camera networks:

Prayagraj, Maha Kumbh 2025: 1,800 AI-enabled cameras out of more than 3,000 tracked crowd density and sent automatic alerts to the ICCC (PIB, January 2025; Hindustan Times).
Delhi, January 2026: the Delhi Police have implemented around 535 ANPR software-enabled projects, with AI-powered ANPR from Vehant flagging stolen and wanted vehicles to the central control room, according to Vehant’s CTO (The Tribune, 31 January 2026).
Ludhiana, May 2026: the ITMS went live at 46 locations and can detect nine types of traffic violation, under an ICCC project with 1,700 cameras at 259 locations (The Indian Express, 6 May 2026).
Mumbai, May 2026: 15,000 new AI-powered cameras under a ₹2,140 crore project, adding facial recognition, speed detection and behavioural analysis to the city surveillance network (The Economic Times/PTI, 19 May 2026).
Chandigarh, August 2026: the police plan to expand the CCTV network from 2,130 to about 4,200 cameras, adding predictive analytics and automated vehicle and person identification (The Tribune, 26 August 2026).
Delhi, September 2026: a Union Cabinet-approved AI traffic network will raise AI-enabled cameras from 209 to about 2,500 across 42 corridors, flagging stolen and hot-listed vehicles and detecting red-light, speed, helmet and seat-belt violations (ThePrint, 30 September 2026).

Earlier, our article on Bhubaneswar’s city-wide surveillance system described one of the first smart city surveillance projects in Odisha.

Getting started: Trisim’s view

Trisim Global Solutions designs and integrates control rooms and surveillance systems. Sourish Dey, Trisim’s Director and Co-Founder, was directly involved in setting up the ICCCs for Bhubaneswar (from 2017) and Agartala.

In Trisim’s view, the most common mistake in city surveillance is buying cameras and analytics licences first and thinking about operations later. Projects work better when they start with the questions the police and traffic teams actually need answered, then choose an open VMS, decide what runs on Local Processing Units and what runs centrally, plan storage and privacy rules, and design the SOPs that tell operators what to do with each alert. Analytics that nobody acts on is just another screen. aigence, an AI video analytics platform, describes itself as a secure-by-design video surveillance suite that combines video management software with video analytics, with ONVIF camera integration and integration with building management sensors and fire alarm detectors.

Looking for smarter, secure-by-design video surveillance? Explore aigence AI Video Analytics.