A headcount during a factory evacuation is not paperwork. It is the number that says whether anybody is still inside, and until it is right nobody can be sent to look. On a working plant that number has to be assembled from 5 separate registers, which is why it arrives after the decision needed it.
Surveyed, installed and commissioned by our own engineers.
Three deployment models, and nobody is subcontracted.
They trust VB
01 In their words
Nobody types these into a search box. They are what an evacuation sounds like from the inside, and none of them is a failure of a person.
I could not tell the fire officer how many people were still inside.
Every record in the building was built for wages, for gate control or for an audit six months later. None of them was ever asked to answer in thirty seconds, standing outside, with a fire officer waiting.
The count at each point is live from the moment the alarm sounds.
By the time we finished counting, the answer was no use to anyone.
Counting by sheet takes as long as it takes. Four marshals, seven contractor supervisors, and a walk back to the table. The number is honest and it arrives after the decision that needed it.
The count arrives while the evacuation is still running.
We nearly sent a team back in for a man who had already gone home.
Exit is not recorded the way entry is, so the register said he was inside. The people who were about to go back in were doing the right thing with the information they had.
The count comes from who is standing at the point, not from the gate.
Every minute we spend counting is a minute nobody is searching.
The marshals who count are the same people who would search. Two urgent jobs, one set of hands, and the counting has to finish first. That is arithmetic, not willingness.
Counting stops being a job that somebody has to stop and do.
Every one of those is time spent not knowing. The rest of this page is the record that measures it.
02 Installed and running
The counting in this page is not a prototype. It runs on the estate below, on the cameras those plants already owned.
100+
Plants and sites where VijAI modules are running
18,200+
Unsafe acts recorded, timestamped and located
0
New cameras required to start
Source · VijAI deployment register, VB Group, September 2026. Site definition and observation period are stated in full in the statistics register.
03 The format
Ten fields, in the order they are completed. Field five is the one most records leave out, and without it none of the others mean anything.
Reference, date, and the exact clock time the alarm was raised. Not "morning".
Shift, announced or unannounced, light and weather. A drill at 11 a.m. that everybody expected proves something different from one at shift change.
Named, and the clause of the emergency plan it tested. "Fire drill" leaves most of the plan untested and the record cannot show which part.
By area, with the muster point each one reports to. Never "whole plant".
The denominator, register by register. Employees, contract labour, drivers, visitors, open work permits. Without this the record has no roll call, only a headcount.
Headcount by point and by group, and the names of anyone expected and not present. A shortfall without names costs the next twenty minutes.
Alarm audible in every area, areas cleared, all points reporting, and roll call closed. Four clock times, not one.
The number outstanding, how each person was located, and the time the figure reached zero. Blank is not the same as zero.
Each with an owner, a target date and the evidence of closure. Open findings carry forward onto the next record so the list cannot quietly disappear.
Safety officer prepares, participating department heads countersign, the occupier approves. Each contractor supervisor signs for their own number.
One worked example. No plant, sector or state is named on this one, because a solvent store plus a state narrows to very few plants.
Yes. No form, no email address, no download wall. Lift it into your own EHS document and change whatever does not fit your site.
EVACUATION DRILL RECORD · FIELD FORMAT 01 IDENTIFICATION reference, date, exact clock time the alarm was raised 02 CONDITIONS shift, announced or unannounced, light, weather 03 SCENARIO what was rehearsed, and the plan clause it tested 04 AREAS by area, with the muster point each reports to 05 POPULATION AT ALARM the denominator, register by register. Five sources 06 MUSTER RETURNS count by point and group, plus NAMES not present 07 TIMINGS audible / cleared / all points reporting / ROLL CALL CLOSED 08 UNACCOUNTED number outstanding, how located, time it reached zero 09 FINDINGS owner, target date, closure evidence. Open ones carry forward 10 APPROVAL safety officer, department heads, occupier, contractor signatures
The numbering is ours and yours will differ. What matters is that field 05 and the closure line survive the translation, because those two are what an inspector and a fire officer each ask for.
Current against · ISO 7010 safety signs, assembly point E007 · the Manufacture, Storage and Import of Hazardous Chemicals Rules, on-site emergency plan and its rehearsal · the contract labour registers a principal employer holds. Re-verified September 2026.
Because nobody owns the denominator, and four of its five sources belong to different departments.
| Population | Recorded in | Held by | Live at the alarm? |
|---|---|---|---|
| Employees | Attendance or access control | HR | Usually |
| Contract labour | Gate pass and muster roll | The contractor, and security | Rarely |
| Drivers and helpers | Security or weighbridge register | Security | No |
| Visitors and vendors | Visitor register | Reception | Sometimes |
| Shutdown crew | Open work permits | Maintenance | No |
Anything in the last column that is not "usually" is a person nobody will miss for the first twenty minutes. The fix is one line in the emergency plan: a named role pulls all five registers at the moment the alarm sounds. It costs nothing and it is the difference between a headcount and a roll call.
Contract labour is the largest of the five on most Indian sites, and it is the one register the plant does not write itself. Each contractor supervisor counting and signing for their own gang closes more of the gap than any other single change.
This is the only number on the record that has anything to do with a real emergency, and almost no drill report carries it.
Until that number is right, nobody can be sent to look. A search party that goes in on a wrong count is either looking for somebody who drove out twenty minutes ago, or not looking for somebody who is still in a trench. Both are decisions taken with the wrong information, and both were taken by people doing exactly what the plan told them to.
An evacuation record that says "all personnel assembled safely" tells a reader nothing, because nobody was counting against a known number. A record that says 412 recorded, 389 at the points, zero unaccounted at 23 minutes 18 seconds is a different document. It says how long the plant was blind, and that is the interval worth shrinking.
It is also the honest way to set a target. The interval is yours, measured on your own site rather than copied from a standard, and the three findings in the worked example above are the ordinary ways it comes down.
Five returns, and none of them is about effort.
"Approximately 350 personnel assembled" cannot be checked against anything. The record has counted heads and called it a roll call, and the two are not the same document.
The plant does not know the gang. The supervisor does. A number the supervisor has not signed for is an estimate wearing a signature block.
The evacuation time is the easy number and the least useful one. Without the closure time the record cannot show how long the plant went without knowing whether anyone was still inside.
A drill that found nothing either tested nothing or was not observed. Three findings with owners and dates is a healthy record; none is a warning.
A point with no marshal produces no return, and it hides anyone who went there. Either staff it on every shift or take it off the plan and remove the sign.
It shortens the gap between the alarm and a count you can act on. It does not give you the names, and this page will not pretend otherwise.
VijAI reads the feed from cameras already covering a muster point and reports how many people are standing in it, continuously, from the moment an alarm is declared. The site main controller watches the totals rise instead of waiting for four marshals to walk their sheets to a table, so the shortfall is visible while there is still something to do about it. Because it counts people rather than reading a token, it counts the driver and the visitor the same way it counts an operator.
It does not produce names, and it does not replace any of the five registers in field five. The denominator still comes from your own records and the named absentees still come from the marshals and the contractor supervisors. What changes is how quickly the difference between the two becomes visible.
Deployment is on-premise, hybrid or cloud, and most Indian plants choose on-premise, so the video and the counts stay inside the plant network. Where a muster point has no camera covering it, that gap is settled at survey before anything else is discussed.
This page is the record. The two below are the module and the wider set.
Occupancy and muster counting on the cameras a plant already has.
AcrossThe rest of the vigilance setThe other detection modules that run on the same installed cameras.
NextPut it to our engineering teamTell us your muster point layout and the last drill you ran.
Sources · ISO 7010, graphical symbols and safety signs, assembly point · the Manufacture, Storage and Import of Hazardous Chemicals Rules, on-site emergency plan and its rehearsal · Annual Survey of Industries, share of contract workers in Indian factories. Named, not linked.
04 How it runs
The first question IT asks is where the video goes. All three answers are here, and detection by default with identification optional applies to every one of them.
| Model | Where the video is processed | Where it fits |
|---|---|---|
| On-premise edge | Entirely inside the plant. Nothing leaves your network. | Licensed areas, sites with a data-residency rule, and plants with no dependable outbound link. |
| HybridRecommended | Detection at the edge, dashboards and reporting hosted. Footage stays on site. Only events leave. | Most plants. Video stays inside the boundary and head office still gets one view across every site. |
| Cloud | Processing and storage hosted. | Single sites, short pilots, and estates already running their VMS this way. |
The model is chosen in the survey, against your network and your data rules. It is not a price tier.
05 Who installs it
Survey, engineering, installation, commissioning, calibration, training and support are all done by VB people. When a camera angle is wrong on a Sunday night, the person who fixes it is the person who chose it.
06 Before your next drill
A headcount counts who arrived. A roll call reconciles who arrived against who was recorded on site, and closes only when the difference is zero and every name is accounted for.
By giving each contractor supervisor a sheet for their own gang and having them sign for the number at the muster point. The gate pass record gives the expected number, the supervisor gives the present number.
Start the sheet and write the time the first person arrives. Counting can wait a minute. The clock cannot be reconstructed afterwards.
There is no published figure to measure against. What matters is that until it closes, nobody can be sent to look for anyone still inside. Plants that time it honestly usually find it takes longer than the evacuation did.
Yes. They are inside the fence and they belong to no muster roll, so a record that omits them has closed on an incomplete population.
07 The next step
A survey tells you which of your muster points your existing cameras already cover. Then one drill, counted from those cameras, and you find out how long your plant currently goes without knowing whether anybody is still inside.
Send an enquiry
