1 month. That is the deployment timeline a Busan-based manufacturer achieved going from signed contract to live AI safety monitoring on its existing CCTV infrastructure. No new cameras. No civil works. The ONVIF auto-recognition software discovered every camera on the network and began running detection — PPE, zone intrusion, man-down, fire — within days of on-site setup.
That deployment shows what the Singapore permit-to-work discussion has not yet addressed: verification. Singapore's Ministry of Manpower (MOM) lists electronic permit-to-work (e-PTW) systems and video analytics as separate Workplace Safety and Health (WSH) technologies on its industry guidance. They are listed separately because they are procured separately, evaluated separately, and typically deployed by different teams. But the operational logic binds them: the e-PTW system records that work is authorised; the CCTV AI verifies that the permit's conditions — PPE worn, zone respected, fire watch present, no unauthorised intrusion — are being honoured in real time.
The e-PTW is a claim. Verification makes it true. The join between these two MOM-listed technologies is the actual operational win, and it is largely unpublished. This post explains how the integration works, where the compliance logic sits, and what the implementation looks like on a site that already has ONVIF-compliant CCTV.
What e-PTW does and what it cannot do
An electronic permit-to-work system replaces paper-based work authorisation with a digital workflow. The permit is issued, approved, and closed through a software platform. Roles — issuing authority, performing authority, area authority — are assigned and tracked with timestamps. Permit conditions are recorded: PPE requirements, isolation points, atmospheric test results, time windows for work, standby personnel assigned. The audit trail is clean, searchable, and exportable.
MOM's WSH Council recognises e-PTW as a control measure that strengthens the work-authorisation process. The benefits over paper are real: no lost permits, no illegible handwriting, no ambiguity about approval sequence, automatic expiry enforcement, and a centralised record for MOM inspections.
The limitation is structural. An e-PTW system records what is supposed to happen. It has no mechanism to verify that what was recorded is what actually happened at the work face. The permit says a fire watch must be stationed at the hot-work location. It says full PPE — helmet, gloves, face shield — must be worn throughout. It says no other personnel may enter the hot-work zone while the permit is active. All of that is registered in the digital record.
Whether the fire watch is actually standing there at 2:15 pm — or stepped away to take a call — the e-PTW system does not know. Whether the worker who entered the zone at 2:40 pm was authorised or not, the system does not know. Whether PPE was removed after the supervisor signed off and walked away, the system does not know.
This is not a gap the e-PTW vendors engineered out of carelessness. It is an inherent boundary: software that manages workflow cannot observe the physical world. That is what cameras are for.
What CCTV AI adds to the permit layer
AI safety monitoring on CCTV does the physical verification that the e-PTW system cannot do. The detection classes relevant to permit-to-work conditions are:
- PPE detection: identifies presence or absence of helmet, vest, gloves, harness, and face shield on individuals within the camera frame at the work zone
- Zone intrusion: triggers an alert when a person enters a defined perimeter — the hot-work zone, confined-space entry point, or isolation boundary — outside the authorised time window or without authorisation
- Person presence: confirms that a required attendant (fire watch, standby person) is stationed at the designated position throughout the permit window
- Fire and smoke detection: visual detection of smoke, flame, or thermal anomaly within the monitored zone
- Man-down detection: identifies a person who has been motionless on the ground beyond a configurable time threshold
Each of these maps to a condition that appears on a standard permit to work. The permit specifies the condition. The camera verifies it is being met.
The integration logic is straightforward. The e-PTW system holds the permit record: what conditions are required, what the work window is, who is authorised to be in the zone. The CCTV AI platform receives the zone boundaries and time window and monitors against them in real time. When the camera detects a condition breach — an unauthorised person in the zone, a fire watch absent for more than a configurable interval, PPE removed during active work — it sends an alert to the safety officer and permit issuer immediately. The event is timestamped and logged in the monitoring platform.
The e-PTW and the CCTV AI platform do not need deep API integration to deliver the core value. Zone boundaries can be configured in the camera platform once per permit type. Time windows can be set manually or fed in via a lightweight data connection to the permit system. The verification layer runs independently and logs events that can be cross-referenced against the permit audit trail.
Permit condition to camera rule: a mapping table
The table below maps standard permit-to-work conditions to the detection rules that verify them. It is structured for safety managers evaluating where camera AI fits their existing permit workflow. This is an implementation reference, not a regulatory sign-off — confirm specific compliance obligations with your WSH officer.
| Permit condition | Verification method | Detection rule | Alert on breach |
|---|---|---|---|
| Full PPE worn throughout work window | Camera at work-zone entry and interior | PPE class detection — helmet, vest, gloves, face shield | PPE absent or removed during active permit |
| No unauthorised personnel in zone | Camera covering zone perimeter | Zone-intrusion detection against authorised-person list | Unrecognised person enters zone |
| Fire watch stationed at hot-work location | Camera with field-of-view of fire watch position | Person-presence monitoring at designated coordinate | Fire watch absent from position >5 minutes |
| No simultaneous incompatible work in zone | Camera covering adjacent work areas | Multi-zone intrusion correlation | Conflicting activity detected in adjacent zone |
| Work confined to permit time window | Time-window rule set from e-PTW | Activity detection outside authorised window | Any detected movement in zone after permit expiry |
| Fire or smoke — immediate stop-work | Camera covering work zone | Fire and smoke visual detection | Alert on any smoke or flame detection |
| Standby person present at confined-space entry | Camera at entry point | Person-presence monitoring at entry coordinate | Standby absent from position during active entry |
| Biometric monitoring for high-risk entry | Smartband worn by entrant | Heart rate, SpO2, skin temperature vs threshold | Physiological alert — SpO2 <94% or HR >130 bpm |
| Permit closed on work completion | Automated activity-end signal | No active detection in zone after permit close | Lingering activity after permit expiry |
| Atmospheric event — stop work | Camera covering zone | Fire/smoke detection + smartband environmental alert | Smoke, flame, or physiological anomaly detected |
Two conditions remain outside camera verification scope regardless of technology:
- Atmospheric gas testing: oxygen levels, flammable gas concentration, and toxic gas readings require physical instrumentation by a competent person before and during confined-space entry. No camera-based system substitutes for gas detection.
- Rescue execution: when an alert is triggered, a trained standby person or emergency responder must physically execute the rescue. The monitoring platform raises the alert — it does not dispatch a rescue team automatically.
These limits are material. Any solution presented as eliminating the competent-person or standby-person obligation is misrepresenting what the WSH framework requires.
Integration architecture: how the two systems connect
The architecture does not require replacing either system. It adds a verification layer between the e-PTW workflow and the physical work site.
Layer 1: e-PTW platform (existing) Manages work authorisation workflow. Outputs: permit record with conditions, authorised personnel, time window, zone designation, required PPE, assigned standby or fire watch. The e-PTW system continues operating as-is.
Layer 2: ONVIF camera network (existing) IP cameras already installed on the site — across work zones, entry points, confined-space locations. If ONVIF-compliant (most commercial IP cameras installed after 2012 are), they are immediately usable. HyperQ AI Safety auto-recognises ONVIF cameras on the local network during setup — no per-camera credential configuration, no vendor-specific SDK, no cabling changes.
Layer 3: HyperQ AI Safety (new software layer) Runs on a local server or edge device. Pulls live video streams from ONVIF cameras. Runs detection model inference on each stream in real time — sub-second alert latency on standard hardware. Detection classes active per zone are configured to match permit conditions for that zone type. Alert routing goes to the safety officer dashboard and mobile notification.
Layer 4: Data bridge (lightweight, configurable) Zone boundaries and time windows from the e-PTW record are input to the camera platform — either manually for stable recurring permit types (hot-work zones, confined-space entry points, electrical isolation boundaries) or via API connection to the e-PTW system for dynamic permit issuance. The bridge does not require the e-PTW vendor's cooperation for the manual configuration path; it requires it only for real-time permit-triggered rule activation.
Layer 5: Cross-reference audit trail Events logged by the CCTV AI platform — PPE breach at 14:17, zone intrusion at 15:03, fire watch absent 14:52–15:01 — carry timestamps that can be cross-referenced against the e-PTW permit record. The combined audit trail is stronger than either system alone for MOM inspection purposes and incident investigation.
The deployment sequence: ONVIF camera discovery (hours), zone configuration per permit type (1–2 days), alert routing and dashboard setup (1 day), operator training (1 day). Software from $10,000. Hardware cost is zero if existing cameras cover the required zones at sufficient resolution and angle. Where coverage gaps exist, ONVIF-compatible supplemental cameras are available at $420–$2,250 per unit — significantly below the cost of a purpose-built hardware-locked safety system.
Why the join matters more than either system alone
A site running e-PTW without CCTV AI verification has clean paperwork and no physical ground truth. Whether the fire watch was actually standing there, whether PPE came off after the supervisor left — none of that is knowable from the permit record.
A site running CCTV AI without e-PTW integration has real-time alerts but no permit context. The camera detects a PPE absence, but the platform does not know whether the person is in an active permitted zone or an idle area. Alert volume is higher; signal-to-noise is lower.
The joined system has both. The e-PTW defines the authorisation boundary. The CCTV AI monitors against it. An alert means: this condition was required by an active permit, and the camera shows it is not being met. The operator has enough context to act immediately rather than investigate first.
For Singapore WSH compliance, the combined audit trail is materially stronger. MOM's inspection focus has shifted toward technology-backed monitoring for high-risk work categories. An incident investigation after a hot-work fire or confined-space fatality routinely examines whether the site had monitoring capable of detecting precursor conditions. A permit record showing PPE was required, paired with a CCTV log showing PPE was removed 20 minutes before the incident, is a very different evidential position than a permit record with no physical verification.
Where this integration is already relevant in Singapore
MOM's WSH Council has specifically named e-PTW and video analytics as technologies under its industry guidance for manufacturing, construction, and oil-and-gas sectors. The SG WSH framework's emphasis on documented control measures for high-risk work categories creates direct procurement pressure on site safety managers to close the verification gap.
The hot-work permit is the highest-frequency high-risk permit type on most industrial sites. Confined-space entry is the highest-severity category. Both generate the clearest ROI case for camera-based condition verification: the number of permits issued per month, the frequency of non-compliance events that manual supervision misses, and the cost of a single WSH incident (regulatory, operational, reputational) make the software investment straightforward to justify.
The CDM 2024 compliance obligations in neighbouring Malaysia follow a parallel trajectory: regulators across the region are moving toward documented technology-backed evidence for high-risk work categories, not procedural checklists alone.
Limitations: where the join does not close the gap
The integration makes sense only if cameras already cover the permit zones at usable angles and resolution. If the existing camera estate was installed for perimeter security and does not see work zone interiors, the verification problem is a hardware-placement problem, not a software one — and adding AI inference does not solve it.
Camera positioning is the first constraint. Existing cameras installed for perimeter security or general surveillance may not cover work zones at the angles or resolution needed for PPE detection or person-presence monitoring. A fire watch stationed at a column 15 metres from the nearest camera, partially obscured, will produce unreliable detection. A coverage audit against permit-zone locations is necessary before any software decision.
Dynamic permit zones are harder than fixed zones. For permit types where the work zone shifts daily — scaffold erection, road-break permits — static zone configuration in the camera platform requires daily updates or API synchronisation with the e-PTW system. Fixed recurring zones (confined-space entry points, hot-work bays, electrical switchrooms) are easy to configure once. Variable zones require a more active integration.
The e-PTW vendor's cooperation affects real-time rule activation. The manual configuration path (safety team updates zone and time-window rules per permit) works without any e-PTW vendor involvement. The automated path — where a permit issued in the e-PTW system instantly activates the corresponding camera rules — requires either an API connection or a lightweight middleware layer. Most major e-PTW platforms have REST APIs; integration complexity is manageable but not zero.
Resolution and lighting in older cameras. PPE detection and person-presence monitoring require adequate pixel density over the monitored zone. Cameras older than 5–7 years running at 720p or below, placed at distance, may not resolve PPE classes reliably. Low-light environments require IR cameras ($650/unit). The platform will flag low-confidence detections rather than suppress them — which is the right behaviour, but does require some camera upgrades on older CCTV estates.
Frequently asked questions
Does camera-based verification satisfy the permit-to-work documentation requirement under Singapore WSH?
The CCTV AI audit trail supplements the e-PTW record — it does not replace it. The permit record remains the primary compliance document. Camera logs provide corroborating evidence of condition compliance, valuable for MOM inspections and incident investigations.
Does HyperQ AI Safety integrate with specific e-PTW platforms?
The platform can ingest zone boundary and time-window data from most e-PTW systems via REST API or manual configuration. Contact the team with your current e-PTW platform details for a compatibility assessment.
What happens if the e-PTW system is down during an active permit?
Camera monitoring operates independently of the e-PTW system. Detection and alerting continue on the last-configured zone rules. The outage does not disable the verification layer, though new permits issued during the outage would not automatically activate corresponding camera rules until the connection is restored.
Can the system distinguish between authorised and unauthorised personnel entering a zone?
Zone-intrusion detection identifies any person entering a defined perimeter. Person identification against an authorised list requires integration with an access-control or ID system. Without that integration, any intrusion into a zone during an active permit window triggers an alert for the operator to verify. For most permit contexts, any unannounced entry into an active hot-work or confined-space zone is a priority alert regardless of who the individual is.
How long does it take to configure camera rules for a new permit type?
Zone boundary drawing and detection-class configuration runs 15–30 minutes per camera view. For a site with 5 permit zone types and 20 cameras, initial configuration is a 1–2 day exercise. Recurring permit zones are configured once and remain active.
Does the platform work for night shifts and low-light conditions?
Standard visual detection requires adequate lighting. For low-light work zones, IR cameras ($650/unit) extend coverage. Thermal cameras ($2,250/unit) detect heat signature independent of visible light and are relevant for hot-work permit monitoring. Both types are supported through ONVIF.
The permit is a claim. Verification makes it true.
The e-PTW record says hot work is authorised, full PPE is required, and a fire watch must be stationed at the location. The CCTV AI platform watches whether those conditions are being met in real time. When the fire watch steps away, the camera knows. When PPE comes off after the supervisor signs and leaves, the camera knows. When an unauthorised person enters the active hot-work zone, the camera knows — and raises the alert immediately rather than at the next manual audit.
MOM lists e-PTW and video analytics separately; the value is in joining them. HyperQ AI Safety has the technical specification, detection class list, and ONVIF compatibility details for evaluating coverage against your permit zones.
Software starts from $10,000. If your existing cameras cover the permit zones at sufficient resolution and angle, there is no hardware procurement. ROI payback runs 11–18 months on sites where the integration replaces manual audit processes and reduces WSH incident exposure.
Map your active permit zones against your existing CCTV coverage and send us the layout. We will identify which permit conditions can be verified on existing cameras today, flag any coverage gaps, and return a zone-by-zone verification assessment within 5 business days — no contract required until the coverage spec is confirmed against your actual site plan. Submit your permit zone layout at apac.hypernology.net/contact.