Wall & Fence Climbing Detection: Perimeter Deterrence
A fence is a statement of intent, not a barrier. The intruder who respects it was never your problem; the one who treats it as a ladder is. What makes climbing hard to catch is its shape — a hand on the coping, a scramble, a drop on the far side, all over in four seconds. Most perimeter analytics are built around presence in a zone or movement across a line, and a person in contact with a vertical structure for three seconds is neither.
Wall climbing detection is written for that shape of event. It watches the structure itself — the top of a wall, a fence line, a gate, a shutter, a railing — and raises an alert when a person transitions from below that line to above it. The event fires while the attempt is still in progress, which is the difference between a deterrence tool and a forensic one.
Why Standard Perimeter Rules Miss It
Two rules carry most deployments and both have blind spots here. A virtual tripwire fires when a track crosses a drawn line; a climbing intruder crosses at the coping, usually where nobody drew one, and where a line does exist the crossing lasts under a second at a shallow angle that tune-out thresholds discard. An intrusion zone fires when someone lingers inside a region — but the climb happens on the boundary, and by the time he is inside, the intervention window has closed.
The sensitivity compromise every operator makes is the other half. Wind-blown vegetation and sweeping headlights generate overnight noise, and the standard response is to lower sensitivity until the false alarms stop. What gets filtered out first are brief, partly occluded events at the fence line — the failure mode our guide to reducing false alarms describes. The detect-assess-respond model the Security Industry Association sets out only holds if detection sees the attempt at all.
What the Model Actually Looks For
Person detection proposes a candidate and a tracker holds its identity across frames — the foundation supporting object detection on the platform. A band annotated along the top of the structure defines the target edge. Then a temporal test: the bounding box must move from below the band to above it, or stay in a climbing posture, for longer than the configured minimum. That minimum sits between one and three seconds — long enough to reject a bird, short enough to alert while he is still on the wall. Because the test is a transition rather than a position, leaning on the outside does not trigger it — the movement over does.
Siting and Camera Geometry
Climbing is a side-on problem. A camera looking square across a fence sees the intruder foreshortened and often hidden by the structure. A camera sighted along the fence line, mounted four to six metres up, sees the entire profile of the climb. That one change in geometry usually matters more than any parameter.
Overlap consecutive cameras by roughly a third of their field of view, so a climb cannot happen unseen at the join. Keep the sun out of the frame at dawn and dusk, or accept the backlit window and compensate with infrared. And annotate the band on the real frame rather than guessing a height: it belongs above what a tall person can reach.
From Alert to Response
An alert is worth exactly as much as what follows it. The outputs that earn their keep are a snapshot plus a ten-second clip pushed to whoever is on duty, a trigger to a talk-down speaker or strobe covering that section of fence, and a log entry with camera, timestamp and clip reference. The deterrence effect comes from immediacy — an audio challenge while he is still on the wall ends a large share of attempts before anything is damaged.
Climbing detection rarely runs alone: it sits alongside intrusion detection for the ground inside the line, line crossing detection for gates and aprons, and the broader perimeter security rules. Where the boundary is a drawn line rather than a wall, virtual fence configuration applies directly. Cameras aimed at a fence are worth protecting from being turned or sprayed: tampering detection covers that.
This is applied end to end in school security system.
For the wider picture around this, start from the gated entrance rules.
Frequently Asked Questions
Can it tell climbing apart from someone standing next to the fence?
Yes. The rule keys on a vertical transition across the band, not on proximity. Standing or leaning on the safe side produces no event. We tune the posture threshold against your footage during commissioning.
Does it work at night, in rain and in snow?
Night works with infrared cameras and modest lighting; rain is handled by the tracker. Heavy snow that buries a low wall is a siting issue, since the band no longer matches the real top edge — re-annotating takes five minutes.
What about hedges, trees or uneven ground along the line?
Vegetation growing through the band is the most common source of drift: either keep the band clear of foliage or accept a seasonal re-annotation. Uneven ground is handled by drawing the band as a polyline following the actual top of the structure.
Do we need to replace cameras or send video to the cloud?
No. The rule runs on your existing streams on an appliance inside your network, from 2 to 128 channels and 1 to 256 TOPS, with no per-camera licence and no video leaving your site. Where a stretch of fence has no coverage, one IP camera at the right angle is enough.
Tell Us Your Scenario
Every site draws its zones differently, and the difference between a rule that works and a rule that gets switched off is usually thirty minutes of configuration. Send us your camera layout and the behaviour you need to catch, and we will tell you which of the 198+ algorithms fit and what accuracy to expect on your footage. Tell Us Your Scenario — or start with a Starter Kit preloaded with tested configurations, running on your own cameras in a single afternoon.