Harnesses and guardrails are usually on site. The danger is the two-minute task when nobody ties off — and nobody is watching.
By the Sentrick Industrial team · Published September 27, 2026
Falls from height are among the most serious hazards on construction sites, in warehouses, and across maintenance work — and fall protection has consistently appeared at or near the top of OSHA's annual list of most frequently cited standards. The rules are well understood. The challenge is that work at height happens in brief, scattered moments across a large site, and no supervisor can watch every ladder, platform edge, and roof opening at once. This article explains where fall protection programs typically break down and how AI vision can help catch unsafe work at height while there is still time to intervene.
Most fall incidents are not caused by a lack of equipment. Harnesses, guardrails, and anchor points are usually on site. The gaps tend to be behavioral and situational: a worker skips tying off for a “quick” two-minute task, a guardrail is removed to move material and never replaced, or a ladder is set up at the wrong angle on uneven ground. These moments are short, which makes them easy to miss during periodic walk-throughs and easy to rationalize in the moment.
Traditional controls — training, permits, toolbox talks, and inspections — remain essential. But they rely on people noticing and correcting a problem in real time, and on a busy site attention is the scarcest resource.
Computer vision running on existing or dedicated site cameras can be configured to recognize specific conditions associated with work at height. Typical use cases include:
The goal is not to replace a competent person's judgment. It is to make sure that person hears about a risky situation in seconds rather than discovering it at the end of the shift — or in an incident report.
A simple detector that flags “person near edge” will fire constantly on a construction site, and alerts that fire constantly get ignored. A more useful approach looks at behavior over time and in context: how long someone has been in the zone, whether the activity matches a permitted task, whether a harness was visible when the person entered, and whether the pattern is unusual for that area and shift. Filtering on context is what turns raw detections into a short list of situations worth a supervisor's attention.
Speed matters more than volume. An effective setup routes a high-priority fall-risk alert to the nearest supervisor or site safety lead immediately, with a snapshot so they can judge the situation at a glance. Lower-priority observations — for example, a recurring pattern of ladders set up in one area — are better handled as daily or weekly summaries that feed toolbox talks and planning.
Over time, anonymized patterns reveal where fall risk concentrates: specific floors or phases of a project, times of day when crews are rushed, or tasks where tying off is consistently difficult. That insight often points to engineering fixes — adding anchor points, installing permanent guardrails, or rethinking the sequence of work — which are more reliable than asking people to remember a rule under pressure.
Workers are more likely to accept camera-based safety tools when the purpose is clearly communicated and the system is designed around hazards rather than individuals. Good practice includes limiting monitoring to defined hazard zones, focusing reports on conditions and trends rather than personal performance, setting clear retention periods, and involving crew leads when defining what the system looks for.
Sentrick Industrial applies behavioral AI to worksite video to detect unsafe conditions — including hazard-zone intrusion and missing PPE — and prioritizes them by risk so supervisors can act while there is still time. It is designed to support an existing fall protection program, not to replace training, permits, or the judgment of a competent person on site.
Fall protection rules are clear; consistent compliance across a busy site is the hard part. AI vision helps by watching the edges, ladders, and lifts continuously, flagging the brief moments when protection lapses, and turning those observations into both immediate interventions and longer-term fixes.