Motion detection is the oldest and cheapest video analytic: it flags when pixels in the frame change from one moment to the next, on the assumption that change means something moved. It runs on almost every camera and recorder, costs almost nothing in compute, and is the classic trigger for motion-based recording — record only when the picture changes, save storage the rest of the time.
Its value is that it is universal and free, and for a quiet indoor scene it works fine. The settings that matter are the detection regions (watch the doorway, ignore the busy road) and the sensitivity threshold, which trade missed events against false ones. As a recording trigger it can cut storage substantially versus continuous recording on low-activity cameras.
The defining pitfall is that pixel change is not the same as a relevant event. Swaying trees, rain, snow, headlights, shadows, and clouds all change pixels and all fire motion detection, so outdoor motion alerts are notoriously noisy — and a too-low sensitivity set to suppress them will also miss the person you cared about. This is exactly the gap AI object detection fills: alerting on "person" or "vehicle" instead of "pixels changed" cuts nuisance alarms by 80–95%. Treat motion detection as a coarse, free first filter, not a reliable alerting analytic on its own.
