Video analytics

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.