The short definition
Line crossing is a directional analytic. The installer draws one or more lines (or open polygons) on the camera scene. Each line is a 2D representation of a 3D plane in the world: a fence, a door threshold, a property line. The engine tracks each detected object's bounding-box centroid frame-to-frame and fires an event when the centroid crosses the line.
Direction sensitivity distinguishes line crossing from zone intrusion. A line has two sides, and the rule can filter on direction, alarming only on the security-relevant one.
Common use patterns
- Fence-line perimeter. Line along the property fence. Inbound crossings alarm, outbound filtered. Pairs with thermal cameras for long perimeters and verified monitoring for response.
- Building dock-door tracking. Line at each dock-door threshold for usage analytics; pairs with LPR for vehicle entry logging.
- Retail entrance counting. Line at the front-door threshold. Inbound and outbound counted separately for occupancy, conversion, and dwell-time.
- Restricted-zone access. Line at a pool deck, playground, or rooftop entrance. Any crossing during restricted hours alarms.
- Tailgating detection support. Line at the door paired with an access-control event. Two crossings within a 5-second window of a single badge swipe flags potential tailgating. See the tailgating entry.
Implementation patterns
Three architectures handle most installs.
- Edge AI. Camera runs object detection plus line-crossing logic on-board. Sends event metadata to the VMS along with the video stream. Lowest latency. Standard on Axis, Hanwha Wisenet 7, Avigilon H5A, Verkada cameras.
- VMS-side analytics. Camera streams raw video to the VMS server, which runs detection and line crossing in the VMS layer. Common in Genetec KiwiVision, Milestone XProtect with analytics, Avigilon Unity. Higher latency but more flexibility.
- Camera-agnostic analytics platforms. Briefcam, Dragonfruit AI, and similar take raw streams from any ONVIF camera and run analytics centrally. Strong for retrofits where existing cameras don't have edge AI.
Configuring line crossing well
Three rules of thumb that prevent the most common configuration mistakes.
- Place the line where bounding-box centroids will cleanly cross. A line drawn at floor level on a low-angle camera will be crossed by feet but the centroid (typically chest-high) won't cross until the person is well inside the zone. Place the line at chest-to-head height in the scene, accounting for camera angle.
- Filter by object class. Specify person, vehicle, or both. Without a class filter, the line will fire on any moving object including animals, leaves, and rain.
- Use direction sensibly. If the legitimate direction (employees walking out at end of day) is the same as the alarm direction, the rule is wrong. Walk the scene, identify normal traffic patterns, and configure direction to ignore them.
When to ask Tec-Tel about line crossing
Line crossing is one of the simplest analytics to deploy and one of the easiest to misconfigure. We'll walk a site, mark up the scenes where line crossing fits, configure direction and class filters, and validate against a sample of real footage. Free scoping call.