Traditional CCTV records video for later review. AI video analytics applies computer vision models to live or recorded streams so teams can find relevant objects, movement and configured conditions without watching every frame.
Detection
An AI model evaluates frames and returns object classes, confidence scores and bounding boxes. Common classes include people, vehicles and objects. Confidence thresholds and scene conditions influence what gets reported.
Tracking
Within-camera tracking associates detections across consecutive frames. This supports direction, dwell time, counting and movement analytics. It is not the same as cross-camera identity tracking.
Context and Vision Policies
Raw detections become operationally useful when combined with zones, lines, schedules, duration and business rules. BAV ONE calls these Vision Policies. A policy evaluates whether a detection, in a specific place, at a specific time, for a specific duration, constitutes an event worth acting on.
Alerts and operational events
When a Vision Policy is satisfied, BAV ONE creates an operational event with permitted evidence, severity and workflow context. AVEKSHA, APIs and webhooks can surface that event to authorized systems.
Deployment
AI video analytics can run at the edge, on-premises, in the cloud or in a hybrid architecture. The right design depends on bandwidth, data-control requirements and operational needs.