OperationsFramework / modelModelAccessible

Video analytics

How can video analytics improve people, teams, or organisational effectiveness?

AccessibleOperationalIndividual2 min read

Contents

Video analytics is the process of extracting information, meaning and insights from video footage.

Video analytics extracts structured information from moving images. In addition to recognising objects or attributes within individual frames, it can track movement and events over time, allowing an organisation to analyse flow, behaviour and unusual situations.

When to use it

Use video analytics when a legitimate business or safety question depends on what happens in a physical space—for example, queue duration, movement through a store, use of a display, congestion, restricted-area access or an emerging hazard.

Video can support identification and security, including applications related to Image Analytics, but biometric recognition requires a particularly strong necessity, accuracy and rights assessment. Non-identifying measures such as counts, dwell time and paths may answer the question with less privacy risk.

Cloud and edge systems can combine multiple camera feeds, detect defined events and alert staff. Continuous monitoring may operate 24/7, but “abnormal” behaviour is context-dependent: a model must be validated, monitored for false alarms and overseen by people rather than assumed to be self-correcting and neutral.

Questions can include:

How many people interact with a product or display, and for how long?

Where do visitors queue, circulate or abandon a journey?

Which operational patterns create delay or safety risk?

Which defined events require a timely security response?

Origins

Video analytics developed from computer vision, pattern recognition and surveillance practice. Earlier CCTV systems mainly recorded evidence for later human review. Digital cameras, cheaper storage, faster processors and machine-learning methods made it practical to detect objects and events automatically, track them across time and search large archives. Contemporary systems increasingly process footage at the edge as well as in central or cloud infrastructure.

What it is

Image analytics examines visual content within a still frame; video analytics adds temporal information. It can estimate direction, speed, dwell, sequence and interaction by following features or objects across frames. Common application groups are:

Identification: recognising or re-identifying an object or person, subject to legal and accuracy constraints.

Behaviour and flow analysis: counting, tracking paths, measuring queues or detecting defined actions.

Situation awareness: identifying events or anomalies that may require human attention.

The output is probabilistic, not a direct reading of intention. Camera angle, lighting, occlusion, crowd density and training data affect performance, so every use case needs a measurable error tolerance and a process for human review.

How to use it

Begin with the decision, not the footage. Define the event or measure, why it is necessary, the minimum data required and who will act on the result. Check authority, notice, consent where applicable, retention, access, security and the rights of employees, customers and bystanders.

Audit existing cameras for coverage and quality. Label representative examples, including difficult conditions, and evaluate the system with false-positive and false-negative rates relevant to the business consequence. Pilot at limited scale, compare results with human observation and monitor drift after deployment.

Where existing CCTV already captures the required scene, analytics may reuse it, but the new analytical purpose can create a new privacy and governance obligation. Prefer aggregation or anonymisation when identity is unnecessary, retain footage only as long as justified and provide an appeal or human-review path for consequential decisions.

Practical examples

A global retailer used existing footage to measure checkout queues. Previously, each store camera retained about a week of recordings before overwriting them. Central analysis allowed the retailer to estimate waiting time, understand movement through aisles and compare which promotions attracted interaction. The operational value came from linking a defined observation to changes in staffing and layout—not from collecting video indefinitely.

Sports-analysis provider Prozone combined feeds around football and hockey pitches to track players. Its system generated more than 10 data-points per second for each player, supporting analysis of distance covered, passes, tackles and off-ball movement. The same temporal logic can support aviation, navigation, industrial operations and emergency response when a defined visual event requires rapid attention.

Top practical tip

Use the least intrusive visual measure that can answer the decision question. Validate it under real lighting, angles and crowd conditions, publish error limits and assign a person to review alerts before consequential action.

Top pitfall

Existing footage is not blanket permission for a new analytical purpose. Do not deploy biometric or employee monitoring simply because it is technically possible. Establish necessity, proportionality, legal authority, security, retention and meaningful human oversight.