What is video analytics?
Video analytics is the automated analysis of video streams to detect events and produce data. Object detection finds people, vehicles, and goods in each frame; object tracking follows them over time; business rules turn the result into an alert (“person in restricted zone”) or a metric (“average queue length by hour”). Most existing security cameras can feed a video analytics system, which is why it is often the first computer vision project a company undertakes.
Processing usually runs on a server at the site or on a device at the camera, with only events and counts sent onward. Streaming raw video to the cloud is costly and often prohibited by privacy policy.
Video analytics in practice
A city public works department points analytics at its existing traffic cameras to count vehicles, bicycles, and pedestrians by direction at twenty intersections. The counts replace manual tube counts and feed signal timing decisions and grant applications. A distribution center uses the same class of software to alert a supervisor when a forklift enters a pedestrian-only zone and to measure dock door occupancy. A retailer measures footfall per department and dwell time at displays without identifying any shopper.
What to watch for
- Counting and zone alerts do not require identifying people, so keep facial recognition out of scope unless there is a clear legal basis for it.
- Post signage and write a retention policy, because video of people is personal data under most privacy regimes.
- Test under night, rain, and glare conditions, which cause most false alerts in outdoor deployments.