Near-Miss Detection: Catching the Accident That Didn't Happen
A near-miss is a close call that hurt no one — the forklift that stopped a metre short, the hand that pulled back before the press cycled. It is a free warning. Cameras with AI detect and log these events automatically, with a short clip, so no worker has to fill a form or admit fault. That turns near-misses into a measurable leading indicator: instead of counting injuries after they happen, you see the risk building and fix the lane, the barrier or the shift before anyone is hurt.
In most Indian factories, near-misses are invisible. Not because they don't happen — they happen every shift — but because almost nobody reports them. Ask a floor supervisor how many close calls occurred last month and you will usually get a blank look or a confident "none." That silence is the problem this article is about.
Why near-misses go unreported in India
The reasons are structural, not personal. A worker who reports a forklift near-miss risks being blamed for being in the wrong place. Production pressure means nobody wants to stop a line to write an incident note. The reporting itself is paperwork in a register no one reads. And admitting a close call can feel like admitting a mistake in front of a supervisor.
The result is a dangerous illusion of safety. A plant can run for months with "zero incidents" on paper while the same blind corner produces a near-strike every week. The first time management hears about that corner is when it finally produces a real injury — and by then the warning has already been paid for in blood.
The safety-pyramid idea — a model, not a law
There is a long-standing concept in industrial safety, usually credited to H.W. Heinrich and later refined by Frank Bird, often drawn as a triangle or pyramid: beneath every serious injury sits a larger number of minor injuries, and beneath those a much larger base of near-misses and unsafe acts. The intuition is that the same hazards that eventually cause a serious injury are throwing off many harmless warnings first.
Treat this as a classic heuristic, not a precise measured ratio. The specific numbers various versions quote have been debated and criticised for decades, and they vary hugely by industry and site. What survives the criticism is the useful, directional idea: a floor generating lots of near-misses is telling you where the serious injury will eventually come from. If you can see the base of the pyramid, you can act before you reach the top — but you can only act on near-misses you actually know about, and manual reporting captures almost none of them.
What cameras can actually detect
This is where AI on cameras changes the economics. The system does not need a worker to notice, remember and report a close call. It watches continuously and flags defined patterns automatically, saving a timestamped clip for each one. Typical detectable events on an Indian shop floor:
| Near-miss type | What the camera detects | The fix it points to |
|---|---|---|
| Forklift–pedestrian close call | A person and a moving truck sharing dangerous ground; distance closing below a threshold | Re-route the walkway, add a barrier or gate at the crossing, mark a keep-clear zone |
| Person in a machine danger zone | A worker crossing into a defined guard/press/robot zone while the machine is live | Fix or add physical guarding, add an interlock, retrain, relocate the control |
| Over-speeding vehicle in an aisle | A forklift or trolley moving faster than a set pace in a shared lane | Enforce a speed limit, add speed bumps or signage, review operator training |
| Worker crossing behind a reversing vehicle | A pedestrian entering the path of a truck backing up | Add a dedicated pedestrian route, a reversing spotter rule, or a wider turning bay |
| Crowding at a pinch point | Too many people clustered at a door, conveyor or narrow gap | Widen or stagger the flow, change break timing, add a second exit |
| Loitering in a no-go zone | A person standing where they should not — under a load, near an edge | Physical barrier, floor marking, or a supervised-access rule |
The point is not the alert in the moment — it is the log over time. Ten forklift near-misses at the same dock aisle in a fortnight is not ten accidents avoided; it is one clear instruction: fix that aisle. Using the same footage to quantify what goes wrong is covered in how to measure the cost of factory downtime.
From a near-miss log to a real fix
The value is only realised if the log drives change. A near-miss dataset lets a plant head do things that a monthly injury count never allows:
- Rank hotspots by frequency. The aisle, machine or crossing throwing the most near-misses gets the first barrier or re-route — evidence-led, not gut-led.
- Change layout with proof. "This lane produced 14 pedestrian close calls last month" is a far stronger case for a capital spend on a barrier than a supervisor's hunch.
- Test whether a fix worked. Re-route the lane, then watch the near-miss count for that zone drop — or not. You get a feedback loop, which manual reporting can never give.
- Spot patterns in time. If near-misses cluster at shift change or the end of a long night shift, the fix may be a shift-pattern change, not a physical one.
Two specific hazard classes have their own deeper treatment: the forklift–pedestrian problem in forklift and pedestrian safety with cameras, and machine zone intrusion in machine guarding and danger-zone detection. Near-miss logging sits on top of both — it is the measurement layer that tells you which of those interventions your floor actually needs.
Honest limits
Camera-based near-miss detection is not a safety oracle, and selling it as one gets people hurt.
- It only sees what it frames. An aisle with no camera, or a hazard hidden behind racking, generates no log. Occlusion, dust, glare and poor light all degrade detection.
- A "near-miss" is a definition, not a fact. The system flags whatever thresholds you set — distance, speed, zone entry. Set them badly and you either drown in noise or miss real events. Tuning to your actual floor traffic matters more than any spec sheet.
- A log is not a control. Detecting close calls does not separate people from trucks. Guarding, barriers, marked lanes, speed limits and training are the controls; the camera measures whether those controls are working.
- Worker trust is not automatic. Introduce the system as a safety tool, not a surveillance-and-punish tool, or you rebuild the exact blame culture that killed reporting in the first place. India's data-protection expectations around worker CCTV apply here.
Where this sits legally
There is no near-miss-specific rule in Indian factory law, but the duty is clear. The Factories Act, 1948 places the responsibility for worker health and safety on the occupier — including providing safe systems of work "so far as is reasonably practicable." A documented near-miss log is direct evidence that you are actively finding and closing hazards, not waiting for injuries. The technical advisory body on occupational safety and health is DGFASLI. Analytics does not discharge the duty — guarding and training do — but a near-miss record is exactly the kind of proactive management a factory inspector wants to see.
Where a plant head should start
Don't wire the whole plant for near-miss logging on day one. Start where the pyramid is widest:
- Pick your two worst conflict points — usually a forklift crossing and a machine danger zone — and log near-misses there first.
- Reuse existing cameras where sightlines already cover those zones; add cameras only where the hazard is unframed.
- Run for a few weeks, then read the log. Let the data name your priority fix rather than deciding in advance.
- Fix, then re-measure. The whole value is the before-and-after count. If the near-miss rate at a zone doesn't fall after your fix, the fix was wrong.
Deciding which zones to watch, and from where, so a camera can actually see the close call it is meant to catch is the survey problem Mama removes: record a short phone walkthrough of your floor and it reads the space — aisles, crossings, machine zones, sightlines — then returns a floor plan and camera-placement plan showing which near-miss zones each camera can genuinely cover, before you buy a single unit.
FAQ
What counts as a near-miss on a factory floor? A close call that caused no injury or damage — a forklift stopping short of a pedestrian, a worker stepping out of a machine zone just before it cycled, a trolley over-speeding past someone. It is a warning that the hazard exists, delivered for free before it hurts anyone.
Why don't Indian factories already track near-misses? Because manual reporting fails under real conditions: workers fear blame, production pressure discourages stopping to file paperwork, and the register no one reads. The result is "zero incidents" on paper while the same hazard recurs every shift.
Does the safety pyramid prove a fixed ratio of near-misses to injuries? No. The Heinrich/Bird pyramid is a classic industrial-safety heuristic — the useful, directional idea that many warnings precede a serious injury. The specific ratios have been widely debated and vary by site; treat it as a model, not a measured law.
Will this catch every close call? No. Cameras only see what they frame, and detection degrades with occlusion, dust and poor light. Near-miss logging is a measurement layer on top of physical controls — barriers, lanes, guarding and training — not a replacement for them.