AI video analytics use case
AI loitering detection for sensitive and idle zones
Short answer
Loitering detection is video analytics that raises an alert when a person remains in a defined area for longer than an allowed time. It flags prolonged presence near ATMs, entrances, perimeters, parking areas or idle zones, which can indicate reconnaissance, an attempted break-in or someone who needs assistance.
How loitering detection works
A person-detection model finds people in each frame and a tracker follows each one, keeping a consistent identity as they move. The system measures how long each tracked person stays within a zone and raises an alert once the dwell time exceeds the limit configured for that zone and schedule.
Because dwell time is measured per person, a busy area with people passing through does not trigger the alert; only someone who stays does.
In OrionAlerts
In OrionAlerts, the AI runtime reads decoded frames from the same video engine that records the cameras, so detection runs on the streams you already record rather than on a parallel feed into a separate analytics appliance. Inference runs on the on-site Nebula OS node with GPU acceleration, and each event is written to the same record as the camera and site it came from. Alerts reach the dashboard, mobile app and email.
What affects reliability
Detection quality depends on the scene as much as the model. These are the factors to check during site survey and pilot.
- Tracking continuity: occlusion by pillars, vehicles or other people can break a track and reset the dwell timer.
- Camera placement: the whole zone should be in view, so people do not leave and re-enter the frame.
- Lighting: low light at night makes people harder to detect and track consistently.
- Time limits: limits should reflect normal behavior in each zone, such as queues or waiting areas.
What you need
No camera replacement is assumed. A typical OrionAlerts deployment needs:
- Existing IP cameras that expose video over ONVIF, RTSP or a supported vendor SDK. Compatibility is confirmed per camera model during solution design.
- An on-site Nebula OS node with GPU acceleration, sized per deployment to the channel count and the detections you run.
- Camera views that actually cover the area to be monitored, with zones and schedules defined for each rule.
- Optional: Orion Cloud for a central view of alerts and health across multiple sites, without uploading raw video.
Where it is used
This use case commonly applies in the industries below. Each industry page covers the risks, detections and modules that matter for that kind of site.
Solution
Which solution it belongs to
Loitering Detection is part of AI Vision & Real-Time Alerts: turn live video into real-time AI alerts for safety, intrusion, PPE, and operational events.
Explore AI Vision & Real-Time AlertsFAQ
Loitering Detection: common questions
How long does someone have to stay before a loitering alert?
That is set per zone by the site team. A short limit suits a sensitive entrance or an ATM lobby at night, while a longer one suits areas where people normally wait, such as a reception or a queue. Schedules let the same zone use different limits at different times of day.
Will people walking past trigger loitering alerts?
No. Loitering detection tracks each person and measures how long they stay inside the zone, so people passing through do not reach the time limit. An alert is raised only when one tracked person stays longer than allowed. Heavy occlusion can break a track and reset the timer, which is why camera placement matters.
Where are loitering alerts delivered in OrionAlerts?
To the OrionAlerts dashboard, mobile app and email, for the teams configured for that site and zone. Each alert carries the camera, site and zone it came from, with the video clip attached, so the responder can see who is there and decide whether to act.
Plan this use case for your sites
We will review your camera views for this use case, scope the deployment for your sites and show it in a live demo.