What is a safety event? How AI defines and captures risk

In traditional safety management, the word "incident" usually means something bad happened — someone got injured, equipment was damaged, or work had to stop. But the most important events for preventing future injuries are the ones where nothing bad happened yet. The near misses. The close calls. The risky interactions that nobody noticed or nobody reported.

In the world of computer vision AI, these moments have a name: safety events. Understanding what a safety event is — and how AI captures them — is key to understanding why this technology is changing the way safety teams operate.

So what exactly is a safety event?

A safety event is any observable interaction or behaviour in the workplace that represents a safety risk — regardless of whether it resulted in injury, damage, or loss.

Different regulatory frameworks use slightly different terminology. In New Zealand, the Health and Safety at Work Act 2015 (HSWA) uses the term "notifiable event" to describe incidents serious enough to require reporting to WorkSafe NZ — including notifiable incidents, which are essentially serious near misses that expose someone to significant risk. In Australia, Safe Work Australia tracks similar categories and reported 188 worker fatalities in 2024, with vehicle incidents accounting for 42% of all deaths — underscoring why detecting vehicle-pedestrian interactions is so critical.

In a computer vision AI platform like inviol, a safety event is detected when the AI identifies a specific combination of objects, movements, and spatial relationships that match a predefined risk scenario. A pedestrian walking within a certain distance of a moving forklift. A vehicle exceeding a speed threshold in a defined zone. A person entering a restricted area during active operations.

The important thing to understand is that a safety event doesn't require something to go wrong. It captures risk as it happens — the moment of exposure — not the outcome. That's what makes it so valuable. It's the data point that traditional safety systems almost never capture, because it depends on someone being present, recognising the risk, and choosing to report it.

How is a safety event different from an incident, a near miss, or an unsafe act?

These terms can get confusing, so let's clarify them. They're all related, but they describe different things.

An incident (or accident) is an event that resulted in actual harm — an injury, illness, or damage. These are the events your EHS team investigates, reports, and tracks as lagging indicators. OSHA requires certain incidents to be recorded and reported in the US, while in New Zealand the HSWA requires PCBUs to notify WorkSafe of notifiable events, and in Australia similar obligations apply under the model WHS laws.

A near miss is an event that could have resulted in harm but didn't, often by luck or quick reaction. OSHA defines a near miss) as "an incident in which no property was damaged and no personal injury was sustained, but where, given a slight shift in time or position, damage or injury easily could have occurred." Near misses are the classic leading indicator — the warning sign before the injury.

An unsafe act is a behaviour that deviates from safe work practices — operating equipment without training, bypassing a safety guard, not wearing required PPE. Research suggests that approximately 88% of workplace accidents are caused by unsafe acts.

An unsafe condition is a physical hazard in the environment — a wet floor, faulty equipment, missing guardrails, poor lighting.

A safety event, as captured by computer vision AI, can encompass any of these. It's a broader category that includes near misses, risky interactions, and unsafe conditions that the AI can visually detect. The term is deliberately inclusive because the AI doesn't distinguish between "a near miss where a pedestrian jumped out of the way" and "a close pass where neither party noticed the risk." Both represent the same safety concern, and both deserve the same coaching attention.

The types of safety events AI can detect

Computer vision AI is trained to recognise specific categories of safety event based on visual patterns. The exact detections vary by platform, but the most common categories in industrial environments include:

Pedestrian-vehicle interactions. A person and a moving forklift, truck, or other mobile plant are detected within a defined proximity. This is the most critical category — vehicle-on-pedestrian incidents are among the leading causes of serious workplace injuries in warehousing and logistics. inviol customers see an average 61% reduction in these interactions after deployment.

Exclusion zone breaches. A person enters a virtual zone that has been defined as restricted — a loading dock during active operations, a machinery exclusion area, or a zone around heavy equipment. The AI detects the breach regardless of whether signage or physical barriers are in place.

Speed events. A vehicle exceeds a defined speed threshold within a specific area. Speeding forklifts are a persistent risk in warehouses, and manual speed checks can only catch a fraction of violations. AI captures every one.

Directional violations. A vehicle travels against the designated traffic flow in a one-way area. These events create head-on collision risk that's particularly dangerous in narrow aisles.

Stopped or stationary risks. A vehicle stops in an area where stopping creates a hazard — blocking a pedestrian crossing, for example, or creating a blind spot at an intersection.

At inviol, the detection categories are focused on the interactions that cause the most serious harm in industrial environments. The system doesn't try to detect everything — it detects the events that matter most for preventing serious injuries and fatalities.

How AI classifies and prioritises safety events

Not all safety events carry equal risk, and a well-designed platform reflects that. When inviol detects a safety event, it classifies the event by type (pedestrian-vehicle interaction, exclusion zone breach, speed event, etc.) and assigns a severity level based on factors like proximity, speed, and the nature of the interaction.

This classification matters because it determines how the event flows through your team's coaching workflow. A high-severity near miss between a pedestrian and a fast-moving forklift will be prioritised for immediate coaching. A low-severity speed event in a low-traffic area might be aggregated into a weekly trend review. The goal is to direct your team's time and attention where it will have the most impact — not to drown them in data.

inviol's reporting tools make this visible through dashboards that show event volumes, severity distributions, trend lines, and heatmaps. Over time, these visualisations reveal not just where events are happening, but how the patterns are changing — which is exactly the kind of leading indicator data that drives proactive safety decisions.

Why capturing safety events changes everything

The traditional safety pyramid (Heinrich's triangle) suggests that for every serious injury, there are roughly 30 minor injuries, 300 near misses, and thousands of unsafe acts and conditions. The logic is straightforward: if you address the events at the bottom of the pyramid, you reduce the likelihood of the events at the top.

The problem has always been capturing those bottom-of-the-pyramid events. Manual near-miss reporting is notoriously unreliable — most near misses go unreported. Safety observations are periodic and subjective. Audits happen on a schedule, not in real time.

Computer vision AI changes this by capturing safety events continuously, objectively, and automatically. It doesn't depend on someone being present, recognising the risk, deciding it's worth reporting, finding the right form, and submitting it before the end of their shift. The AI watches every camera, every minute, and captures every detectable event.

For the first time, safety teams have access to a near-complete picture of what's actually happening on their floor — not a sample, not an estimate, but a continuous stream of real, timestamped, categorised safety data.

From event to action: closing the loop

Of course, capturing safety events is only valuable if it leads to action. A database full of events that nobody reviews is just a more comprehensive record of risk that went unaddressed.

That's why inviol connects every safety event to a coaching workflow. Events are reviewed by supervisors, discussed with team members, and logged as coaching conversations. This creates a complete cycle: the AI detects the event, the supervisor coaches on it, the system tracks whether the behaviour changes, and the data shows whether your interventions are working.

This is the approach that delivers inviol's headline results — an average 67% reduction in safety risk and 42% reduction in incidents over three years. The safety event is the starting point. The coaching conversation is what turns it into change.

Want to see what safety events your cameras are already capturing? Book a demo and we'll show you how inviol detects and classifies safety events on your site — turning invisible risk into visible, coachable data.

Frequently Asked Questions

What is a safety event?

A safety event is any observable interaction or behaviour in the workplace that represents a safety risk, regardless of whether it resulted in injury or damage. In computer vision AI, safety events are detected automatically when the system identifies risky interactions — such as a pedestrian near a moving forklift, an exclusion zone breach, or a vehicle speed violation.

What is the difference between a safety event, a near miss, and an incident?

An incident results in actual harm (injury or damage). A near miss could have resulted in harm but didn't. A safety event is a broader category used in computer vision AI that captures any risky interaction the system detects — including near misses, unsafe acts, and unsafe conditions — regardless of outcome. The key difference is that safety events are captured automatically by AI, while near misses and incidents typically rely on manual reporting.

What types of safety events can AI detect?

Common categories include pedestrian-vehicle interactions (people too close to moving forklifts or trucks), exclusion zone breaches, vehicle speed violations, directional violations, and stationary vehicle hazards. The specific detections depend on the platform, but responsible providers focus on the interactions most likely to cause serious injury.

How does AI prioritise safety events?

Computer vision AI classifies events by type and assigns a severity level based on factors like proximity, speed, and the nature of the interaction. High-severity events (such as a close-proximity pedestrian-forklift near miss) are prioritised for immediate coaching, while lower-severity events may be reviewed as part of weekly trend analysis.

Why are safety events important for preventing injuries?

Safety events represent the base of the safety pyramid — the near misses and risky interactions that precede serious injuries. By capturing these events continuously and automatically, computer vision AI gives safety teams the leading indicator data they need to intervene before someone gets hurt, rather than investigating after the fact.