Video analytics applies software to camera streams to detect defined events such as a person crossing a line, a vehicle stopping or an object left behind. Detection runs in the camera or on a server, and results range from useful to unusable depending on placement, lighting and tuning.
Analytics move attention from watching to reviewing. They are useful only when the number of false alerts stays low enough that someone still trusts the system.
On-camera and server-side
Camera-based analytics keep network traffic small because only events travel, but they are limited by the processor inside the camera. Server-side analytics can combine cameras and heavier models, at the cost of bandwidth and compute.
Why false alarms happen
Rain, shadows, insects near the lens, swaying vegetation and reflective surfaces all produce events. Tuning zones and sensitivity, and matching the lens to the distance, reduces them more than adding filters.