Computer Vision for Safety: A New Lens on EHS
Workplace safety has never lacked rules, inspections, or procedures, yet many risks still go unseen in real-time. A safety officer can only inspect a site a few times a day, and CCTV can record everything but interpret nothing. This is where computer vision is changing EHS monitoring. This guide explores that shift: why context and scenario-based detection matter, how visual data becomes actionable safety intelligence, and where technologies like Edge AI and VLMs could take workplace safety next. It also covers practical considerations – accuracy, alert fatigue, privacy, integration, and measurable outcomes – before closing with real-world viAct case studies.
The goal isn't to replace safety teams, but to give them a continuous, contextual lens on risk.
August 26, 2026
Gary Ng
CEO
In this guide
Why EHS Needs a New Approach to Safety Monitoring?
EHS teams have always relied on observation to keep workplaces safe. Inspections, toolbox talks, audits, CCTV reviews, incident reports, and worker observations all play an important role in identifying hazards. However, many of these methods are inherently periodic or reactive. A site may be inspected at a particular time, an incident may be reviewed after it occurs, or CCTV footage may only be examined when someone already knows what to look for.
The challenge is that workplace risk does not follow an inspection schedule.
Construction sites, warehouses, manufacturing facilities, ports, and other high-risk environments change constantly. Workers move between zones, vehicles interact with pedestrians, equipment starts and stops, materials are lifted, and temporary hazards can appear within minutes. A condition that was safe during a morning inspection may become hazardous later in the day.
Why are traditional EHS inspections not enough for today's complex worksites?
Traditional inspections remain essential, but they provide only a snapshot of site conditions. Human observers cannot realistically watch every activity across a large or continuously operating workplace. Even experienced safety professionals have limited visibility, particularly when multiple activities happen simultaneously.
This creates an observation gap—the difference between what is happening across a workplace and what safety teams are able to observe, record, and act upon.
Computer vision can help close this gap by providing continuous monitoring of predefined safety scenarios. Instead of relying solely on someone noticing a risk during a physical inspection or reviewing hours of footage later, AI can analyse video streams continuously and flag relevant events as they occur.
The objective is not to eliminate inspections or human oversight. Rather, it is to extend the reach of EHS teams so that their expertise can be focused where it is most valuable: understanding risk, investigating events, engaging workers, and taking corrective action.
What is the difference between CCTV monitoring and computer vision for safety?
The difference is less about the camera itself and more about what happens after the footage is captured.
A CCTV system gives EHS teams access to what happened. Computer vision can help organise that visual information around specific safety questions. This changes how teams spend their time: instead of manually scanning footage to find something unusual, they can focus on events that have already been identified as relevant to a defined safety scenario.
EHS Need
CCTV Monitoring
Computer Vision
Know what happened
Provides recorded footage for review
Identifies relevant events within footage
Monitor multiple areas
Requires people to watch or review multiple feeds
AI can analyse multiple camera feeds continuously
Find a specific event
Often requires manual searching through footage
Relevant events can be automatically surfaced
Spot recurring issues
Requires repeated manual review and documentation
Detected events can be structured for trend analysis
Prioritise attention
Depends on the operator noticing an issue
Defined scenarios can be prioritised through automated alerts
Investigate an event
Footage provides visual evidence
AI-generated event information can help direct the investigation
Understand site patterns
Difficult to derive at scale from raw footage
Aggregated events can reveal recurring locations, times or scenarios
Support proactive safety
Primarily provides visibility and evidence
Can support earlier intervention when a defined risk occurs
On a busy construction site, CCTV can provide extensive visual coverage, but finding specific events—such as repeated vehicle–worker proximity incidents—may require hours of manual review.
Computer vision adds an analytical layer that can identify defined safety scenarios and organise relevant events for EHS teams. However, automation alone does not guarantee better results. Camera positioning, lighting, obstructions, and poorly defined scenarios can affect performance.
The goal is therefore not:
CCTV → AI → More Alerts
but:
Video → Relevant Safety Events → Actionable Insights → Better EHS Decisions
Computer vision works best when AI detection is combined with well-defined safety scenarios and human expertise.
What does continuous safety intelligence mean for EHS teams?
Continuous safety intelligence gives EHS teams visibility into workplace risks beyond scheduled inspections. Instead of waiting for an incident or manually reviewing footage, AI can help identify relevant safety events as they happen and highlight recurring patterns.
For EHS teams, this can mean:
-
Fewer blind spots between inspections
-
Earlier visibility into emerging risks
-
Better evidence for investigations
-
Faster identification of recurring safety issues
-
More informed prioritisation of corrective actions
The goal is not to replace safety professionals, but to give them better and more continuous information to make safety decisions.
How Computer Vision Brings Context to Workplace Safety?
Computer vision becomes more useful for EHS when it can consider what is happening, where it is happening, and what other elements are involved—rather than simply detecting individual objects.
A worker, vehicle, or missing PPE item on its own may not indicate an immediate safety risk. Context helps determine whether those elements form a potentially hazardous situation.
For example, detecting a worker without a harness is one observation. Detecting that the same worker is near an unprotected edge creates a more meaningful safety scenario.
This contextual approach allows computer vision systems to focus on relationships and conditions, helping EHS teams receive more relevant safety information instead of isolated detections.
What is the difference between object detection and scenario understanding?
Object detection identifies individual elements in a scene, such as a worker, helmet, vehicle, or safety harness. Scenario understanding goes a step further by interpreting how these elements interact—such as a worker without a harness working near an open edge—to identify whether the situation presents a meaningful safety risk.
Can computer vision understand what is happening in a scene?
Computer vision can go beyond identifying individual objects by analysing their location, movement, proximity, and relationships within a scene. For example, detecting a worker, forklift, and restricted zone separately provides basic information; understanding that the worker has entered the forklift operating zone provides meaningful safety context.
This depends on how the AI is trained and the scenarios it is designed to recognise. In EHS applications, contextual detection can consider who is involved, what they are doing, where they are, and what hazards are nearby. This allows the system to focus on relevant safety situations rather than treating every detected object as a potential risk.
Why does context matter in AI-powered safety monitoring?
Context matters because the same visual observation can represent very different levels of risk depending on the surrounding conditions. A worker without PPE, for example, is not necessarily exposed to the same hazard in every location or activity.
By considering factors such as location, activity, proximity, movement, and nearby hazards, AI can distinguish between an isolated observation and a situation that requires intervention. This helps EHS teams focus on meaningful safety events rather than receiving alerts for every detected condition.
For example, detecting a worker without a harness is one observation. If that worker is performing work near an unprotected edge, the combination of worker + missing harness + work at height + open edge represents a much more significant safety scenario.
The result is more relevant alerts, less alert fatigue, and better prioritisation of safety interventions. Context therefore helps shift AI-powered monitoring from simply detecting what is visible to identifying what actually matters from a safety perspective.
Scenario-based AI & Reduced alert fatigue
Scenario-based AI reduces alert fatigue by looking at multiple conditions together before triggering an alert. Instead of flagging every worker without PPE, for example, the system can consider whether the worker is also in a high-risk location or exposed to a relevant hazard.
This helps EHS teams receive more meaningful alerts and fewer irrelevant notifications. The aim is not simply to reduce the number of alerts, but to improve their relevance so that teams can focus their attention on situations that genuinely require intervention.
Does every PPE violation need an alert?
No. A missing PPE item does not always represent an immediate safety risk. For example, a worker without a harness may be in a designated safe area where fall protection is not required.
Scenario-based AI can consider factors such as location, activity, hazard and proximity before triggering an alert. This helps distinguish a simple PPE observation from a situation where the missing protection creates a genuine safety concern.
How does scenario-based AI reduce false safety alerts?
Scenario-based AI reduces false safety alerts by looking at the complete situation rather than treating one detected condition as a safety violation on its own. Traditional rule-based detection may simply identify that a worker is not wearing a harness and immediately generate an alert. However, the system may not know whether that worker is actually exposed to a fall hazard. Scenario-based detection can consider additional factors such as the worker’s location, activity, proximity to an edge, and surrounding conditions before deciding whether the situation matches a defined risk scenario.
For example, imagine a construction site where a worker is walking across a ground-level area without a safety harness. If harness use is not required in that zone, generating an alert would create unnecessary noise. Now consider the same worker standing near an unprotected edge while performing work at height. The combination of worker + no harness + open edge + relevant work activity represents a much more meaningful safety scenario. By evaluating these conditions together, AI can avoid flagging every isolated PPE observation and instead focus EHS teams on situations that require attention. This helps improve the relevance of alerts and reduce alert fatigue without simply reducing the number of alerts.
How does AI determine whether a detected safety condition represents a real risk?
AI can evaluate a detected condition against the context in which it occurs. Instead of treating every observation as a violation, the system can consider factors such as the worker’s location, activity, nearby hazards, movement, and defined safety zones to determine whether the conditions match a specific risk scenario.
For example, detecting a person inside a vehicle operating area does not automatically mean there is an incident. If the person is crossing through a designated pedestrian route, the situation may be acceptable. If the same person enters an active vehicle exclusion zone while a forklift is approaching, the combination of location, movement, proximity, and activity creates a meaningful safety risk. This contextual approach helps AI distinguish between an observation and an actionable event.
Can AI combine location, activity and proximity before generating an alert?
Yes. This is one of the key advantages of scenario-based AI: it can evaluate multiple conditions together rather than triggering an alert from a single observation. Depending on the safety scenario, the system can consider where a person is, what they are doing, what equipment or hazard is nearby, and how close they are to it.
For example, simply detecting a worker and a forklift in the same area may not indicate a risk. But if the worker enters a defined vehicle exclusion zone while the forklift is moving toward them, the combination of location, activity and proximity can meet the conditions for an alert. This allows AI to focus on the interaction that creates the risk rather than treating every individual detection as an incident.
Proactive Safety: Detecting Risk Before It Escalates
Traditional safety management often relies on incidents and observations to reveal where risks exist. Computer vision can add another layer by identifying near misses, unsafe interactions, and recurring hazardous conditions as they develop. For example, repeated instances of workers and moving vehicles coming into close proximity may indicate a risk in a particular work zone even when no incident has occurred.
The value is not in claiming that AI can predict exactly when an accident will happen. Instead, repeated visual events can provide early signals of changing risk conditions. EHS teams can use these signals to investigate the underlying cause, adjust controls, improve site practices, or intervene before the same situation escalates into an actual incident.
Can computer vision detect near misses?
Yes, computer vision can be configured to identify predefined near-miss scenarios by analysing interactions between people, vehicles, equipment, and hazardous areas. For example, if a pedestrian enters a vehicle exclusion zone while a moving forklift approaches, the system can recognise the unsafe proximity even if no collision occurs.
The important point is that AI is not determining whether an event is legally or formally classified as a “near miss.” It is detecting the visual conditions associated with a potential near miss. These events can then be reviewed by EHS teams, documented, and analysed to identify recurring hazards and opportunities for preventive action.
Can AI identify line-of-fire and unsafe-proximity risks?
Yes. Computer vision can be configured to monitor the spatial relationship between people, moving equipment, and defined hazard zones. For example, it can identify when a worker enters the operating area of a moving forklift, stands too close to a reversing vehicle, or moves into the path of a suspended load.
The key is context and movement, not simply detecting that two objects are present. By considering factors such as distance, direction, movement, and predefined exclusion zones, AI can flag situations where a person may be exposed to a line-of-fire or proximity hazard, allowing EHS teams to intervene before the situation escalates.
Can repeated visual events become leading safety indicators?
Yes. Repeated visual events can contribute to leading safety indicators by showing patterns of unsafe conditions before they result in an incident. Traditional lagging indicators—such as injuries, accidents, lost-time incidents, or property damage—tell an organisation what has already happened. Leading indicators, in contrast, focus on activities and conditions that can provide an earlier view of safety performance, such as safety observations, corrective-action completion, hazard reporting, and near-miss trends. Computer vision can add another source of evidence by continuously capturing predefined safety events across operational areas.
For example, suppose an AI system repeatedly detects pedestrians entering a vehicle exclusion zone around the same loading bay. A single event may simply require an intervention. However, if the pattern continues over several days or weeks, it becomes a more useful safety signal. EHS teams can investigate whether the underlying cause is poor traffic segregation, inadequate signage, changing workflows, congestion, or insufficient pedestrian controls. Similarly, recurring instances of workers entering restricted areas, unsafe proximity to moving equipment, or exposure to open edges can reveal where preventive controls may need strengthening.
The important distinction is that the AI-generated event itself is not automatically a leading indicator. It becomes useful as a leading indicator when organisations define, track, and interpret these events as part of their broader EHS measurement framework. Trends such as near-miss frequency, repeated unsafe conditions, high-risk-zone activity, intervention rates, and recurring violations can complement existing leading indicators and help safety teams identify where action may be needed before an incident occurs.
This creates a more proactive safety cycle:
Detect recurring condition → Analyse the pattern → Investigate the underlying cause → Strengthen controls → Monitor whether the risk decreases
Computer vision therefore does not replace established EHS indicators. It can expand the evidence available to EHS teams, helping them move from measuring only outcomes to understanding the conditions and behaviours that may influence those outcomes.
Turning Workplace Video into Actionable Safety Insights
Video footage has traditionally been valuable for visibility and investigation—showing what happened after an event occurs. Computer vision can give that footage another role: turning visual observations into structured safety information that EHS teams can analyse over time.
Instead of treating every detection as an isolated alert, organisations can look at where events occur, how frequently they happen, what types of risks recur, and whether conditions are improving or worsening. For example, repeated unsafe proximity events around one loading area may indicate a need to review traffic movement or pedestrian controls. Over time, these patterns can help EHS teams prioritise areas that require attention rather than relying only on individual observations.
The result is a shift from video as evidence to video as EHS data. When combined with inspections, incident records, IoT data, and other safety information, computer vision can help create a more complete picture of workplace risk and support better-informed preventive decisions.
What Can EHS Teams Learn From AI-Powered Safety Dashboards?
An AI-powered safety dashboard can bring together data from multiple safety scenarios and turn individual detections into a site-level view of safety performance. Instead of reviewing alerts one by one, EHS teams can use dashboards to understand where risks are concentrated, how safety performance is changing, and which areas may require attention.
A well-designed dashboard can provide:
-
Safety Score: An overall view of safety performance based on defined safety parameters and monitored events.
-
Site-wise Safety Scorecards: Compare safety performance across multiple projects, facilities, zones, or sites.
-
Safety Trend Reports: Track whether safety performance and specific risk categories are improving or deteriorating over time.
-
Risk Heatmaps: Visualise locations where safety events or high-risk activities occur most frequently.
-
Scenario-wise Analysis: Break down events such as unsafe proximity, PPE violations, restricted-area access, or work-at-height risks.
-
Time-based Insights: Identify shifts, days, or operating periods associated with higher risk.
-
Event and Intervention Tracking: Follow detected events through review, intervention, and corrective action.
-
Cross-site Benchmarking: Help organisations identify sites performing well and those requiring additional attention.
For example, a site-wise safety scorecard may show that one facility is consistently performing below the organisation's overall benchmark. A corresponding heatmap could then reveal that most events are concentrated around its loading area. The EHS team can drill further into the trend report to determine whether those events are increasing during particular shifts or operational periods.
This makes the dashboard more than a visual display of AI detections. It becomes a decision-support layer for EHS, helping teams move from “How many alerts did we receive?” to “Where is risk increasing, what patterns are we seeing, and where should we act first?”
How can EHS teams identify high-risk zones using visual data?
Visual data can help EHS teams identify areas where safety events occur more frequently or where multiple risk factors repeatedly overlap. By analysing detections by location, frequency, scenario type, and time, teams can build a clearer picture of which parts of a site may need closer attention.
For example, a safety heatmap may show repeated vehicle–pedestrian proximity events around a particular loading bay. If the same area continues to generate events across different shifts, it becomes a useful signal for investigation. EHS teams can then review factors such as traffic routes, congestion, signage, access controls, or work practices and determine whether additional controls are needed.
This approach helps shift site analysis from simply asking “What incidents occurred?” to also asking “Where are risks repeatedly emerging?” Visual data can therefore support targeted inspections, corrective actions, and more informed allocation of safety resources.
Can organisations use computer vision to compare safety trends across sites?
Yes. When computer vision data from different locations is connected through a centralised safety management platform, organisations can view and compare safety performance across multiple sites from a single interface. Instead of each site maintaining separate AI alerts and reports, centralised management brings events, safety scores, trends, and other insights into one view.
For example, an organisation managing seven sites can compare site-wise safety scorecards, recurring risk categories, safety trends, and high-risk zones to identify where performance is improving and where additional attention may be needed. A centralised platform can also give corporate EHS teams a consistent way to monitor deployments, while individual site teams can focus on the risks relevant to their own operations.
This creates a connected management approach: site-level monitoring feeds into centralized EHS intelligence, helping organisations benchmark performance, identify recurring issues across locations, and make more consistent safety decisions at scale.
Making Safety Monitoring Faster with Edge AI
Edge AI brings AI processing closer to where video is captured, allowing safety events to be analysed with less dependence on sending every video stream to a remote server. This can support faster detection, quicker alerts, and more responsive monitoring, particularly in sites where connectivity, bandwidth, or response time are important.
Why is Edge AI important for real-time safety?
Edge AI processes video close to the camera, reducing the time required to send footage to a remote server, analyse it, and return a result. This can be particularly useful for safety scenarios where seconds matter, such as vehicle–pedestrian proximity, restricted-area entry, or unsafe work near moving equipment. By analysing the scene locally, the system can detect a defined safety condition and initiate an alert without waiting for the complete video stream to travel to a central cloud environment.
For example, consider a worker unexpectedly entering the path of a moving forklift. An edge device can analyse the camera feed locally, recognise the defined worker + moving vehicle + unsafe proximity scenario, and trigger an alert almost immediately. Similarly, at a construction site, Edge AI can detect a worker entering a restricted zone around lifting operations and notify the relevant team while the situation is still developing. This makes Edge AI particularly valuable where low latency, reliable local processing, and rapid intervention are important to safety operations.
Is Edge AI better than cloud AI for safety-critical applications?
Not necessarily. Edge and cloud AI serve different purposes, and many safety systems can benefit from using both. Edge processing is useful for fast, local detection and alerting, while cloud infrastructure can support centralised data storage, dashboards, trend analysis, model management, and multi-site reporting.
Edge AI
CCTV Monitoring
Faster local processing
Centralised processing
Lower dependence on connectivity
Easier multi-site data management
Useful for real-time alerts
Strong for analytics and reporting
Reduces continuous video transmission
Supports large-scale data aggregation
Can continue operating during connectivity interruptions
Enables centralised model and system management
Does processing video at the edge improve privacy and reduce data transmission?
It can. When video is analysed locally, organisations can process the footage at the source and transmit events, metadata, or selected information rather than continuously sending full video streams to the cloud. This can reduce bandwidth requirements and limit how much raw video leaves the site.
However, Edge AI does not automatically guarantee privacy. Organisations still need appropriate data-retention policies, access controls, encryption, camera placement, and governance practices to determine how video and safety data are collected, stored, and used.
Are VLMs and AI Agents an upgrade to traditional computer vision?
VLMs and AI Agents can be seen as an evolution of computer vision rather than a simple replacement for it. Traditional computer vision is highly effective at detecting defined objects, actions, and safety scenarios, while Vision Language Models (VLMs) can add a broader layer of scene understanding by interpreting relationships and context within visual data.
AI Agents can take this further by connecting visual observations with reasoning, workflows, and actions. For example, instead of simply detecting a worker entering a restricted area, an agent could help interpret the event, prioritise its severity, retrieve relevant information, trigger a workflow, and support the EHS team in deciding what needs attention. This creates a progression from “detecting what is visible” to “understanding what it means and what should happen next.”
What are VLMs and how are they different from traditional computer vision?
Vision Language Models (VLMs) combine visual understanding with language-based reasoning. Traditional computer vision is generally trained to identify specific objects, actions, or predefined scenarios—for example, detecting a helmet, forklift, worker, or restricted zone. A VLM can interpret a broader scene and describe what is happening, how different elements relate to each other, and what the situation may mean.
For EHS, this difference can be useful when situations are more complex than a single detection. For example, traditional vision may identify a worker, a ladder, and an open edge separately, while a VLM can help interpret the relationship between these elements and the overall scene context. Rather than replacing scenario-based computer vision, VLMs can add another layer of contextual understanding for situations that are difficult to capture through fixed detection rules alone.
Can AI understand and explain complex safety scenes?
Yes, newer AI approaches such as Vision Language Models (VLMs) can analyse multiple elements in a scene and provide a more contextual interpretation. Instead of only identifying a worker, vehicle, or piece of equipment, the AI can consider how these elements relate to one another and describe the situation in natural language.
For example, on a busy construction site, AI may identify a worker, a moving excavator, a temporary barrier, and a nearby access route and interpret how their positions and activity could create a potential safety concern. This can help EHS teams understand why an event was flagged, rather than receiving only a basic detection label. Human review remains important, particularly for high-consequence safety decisions.
Can EHS teams ask questions about visual safety data in natural language?
Yes. A natural-language interface can make safety analytics easier to explore without requiring EHS teams to navigate multiple reports or manually filter large datasets. An EHS manager could ask questions such as “Which site had the highest number of near-miss events this month?”, “What were the most common risks during night shifts?”, or “Has the safety score improved since corrective actions were introduced?”
The value is in making existing safety information easier to query and interpret. Instead of searching through individual events, teams can ask follow-up questions, compare sites or periods, and quickly identify the information most relevant to a particular safety decision.
What could an AI safety copilot look like?
An AI safety copilot could act as a natural-language layer over an organisation’s existing EHS data, helping safety teams find information, summarise trends, and investigate specific events without manually moving between dashboards, reports, and video records.
For example, an EHS manager could ask, “Show me the top three risk areas across our sites this month” and then follow up with “Why is Site 4 performing worse?” or “What changed after the corrective action?” The copilot could bring together safety scores, event trends, heatmaps, incident records, and visual detections to provide a contextual response and point the user toward the underlying evidence.
The important distinction is that a copilot should support safety decisions, not make them independently. Its role is to reduce the time spent searching, comparing, and interpreting information so EHS professionals can spend more time on investigation, prevention, and action.
Case Studies: Turning AI Alerts into Meaningful EHS Outcomes
The real value of computer vision is not in generating more alerts—it is in helping organisations identify meaningful risks and act on them. Across industries, AI deployments show how scenario-based AI can turn everyday video data into measurable safety improvements, from reducing unsafe proximity and zone intrusions to improving safety scores and reducing manual monitoring effort.
The following case studies highlight how organisations moved from isolated safety observations to continuous, measurable EHS intelligence, while keeping human judgement at the centre of safety decisions.
Are fewer alerts always better?
Not necessarily. The objective of an AI safety system should be relevant alerts, not simply fewer alerts. An alert that identifies a genuine high-risk interaction can be far more valuable than dozens of low-priority notifications.
For example, at a Singapore construction company managing multiple high-rise and infrastructure projects, vision AI was deployed across multiple safety scenarios using existing CCTV infrastructure. The deployment helped the organization achieve a 10× improvement in safety score, while saving more than 7,000 working hours through a shift toward proactive safety management.
This illustrates an important principle: AI performance should be judged by whether alerts help teams identify and address meaningful risks, rather than by the raw number of alerts generated.
How can companies reduce false positives without missing genuine risks?
Reducing false positives requires AI to understand the conditions surrounding an event. Scenario-based detection can combine factors such as location, activity, proximity, and hazard context instead of treating every isolated detection as a violation.
viAct's deployments demonstrate this approach across different environments. For example, its mining case study reports a 70% reduction in proximity risks and a 65% reduction in zone intrusions using dynamic safety zoning. In a U.S. manufacturing plant, AI monitoring of workers entering suspended-load areas resulted in an 82% reduction in unsafe entries within three months.
The broader lesson is that accuracy is not only about how often AI detects something. It is about whether the system can distinguish a meaningful safety scenario from an ordinary workplace event.
How should organizations measure the accuracy and ROI of AI safety monitoring?
Accuracy should be measured alongside operational and EHS outcomes. Useful metrics can include detection accuracy, false-alert rates, response time, repeat violations, safety-score changes, hours saved, and reduction in incidents or high-risk exposure.
The ROI can then be evaluated through measurable outcomes. For example, viAct reports 7,000+ working hours saved in its Singapore construction deployment, while its UAE dairy and beverage case study reports a 40% reduction in hygiene violations and continuous monitoring without additional manpower.
This gives organisations a more realistic way to evaluate AI: Is it improving safety performance, reducing manual effort, and helping teams act earlier?
Which leading indicators can improve through computer vision?
Computer vision can strengthen leading-indicator programs by continuously capturing unsafe conditions and behaviours before they become incidents. Depending on the deployment, these can include near-miss events, unsafe proximity, restricted-zone entries, PPE compliance, repeated violations, response times, and risk concentration by location or shift.
For example, viAct's Hong Kong port deployment reported a 10× improvement in lift-zone safety, while its Saudi construction deployment combined video analytics and smart-watch data to achieve a 63% reduction in on-site medical emergencies associated with heat stress.
These examples show why computer vision can complement traditional EHS indicators: it provides continuous operational evidence of the conditions that precede incidents, allowing teams to identify patterns and intervene earlier rather than relying only on lagging outcomes
How viAct Turns Computer Vision into Actionable Safety Intelligence?
viAct’s approach goes beyond simply putting AI on top of CCTV. Its scenario-based Vision AI is designed to identify safety situations by considering contextual elements rather than detecting isolated objects. Across its deployments, existing CCTV infrastructure can be connected to viHUB and multiple AI modules—including PPE detection, open-edge detection, machinery tracking, confined-space monitoring, and access control—creating a unified view of safety conditions.
What Safety Challenge Did viAct Address?
A recurring challenge across high-risk industries is that manual inspections cannot provide continuous coverage. In viAct's Singapore construction case study, a major contractor was dealing with inconsistent PPE compliance, near misses around heavy machinery, and hazards around open edges and confined spaces. Manual audits and paper-based reporting made safety management reactive, with important information often becoming available only after a risk had already occurred.
How Did Scenario-Based Detection Improve the Monitoring Process?
viAct connected multiple safety scenarios to the organisation's existing CCTV infrastructure, turning cameras into continuous safety monitors. Rather than relying on one generic detection, different modules could monitor specific conditions across the sites, while viHUB centralised the resulting events and safety information into a live dashboard. This allowed EHS teams to see site-wise safety performance, safety trends, and emerging risks without manually reviewing every camera feed.
The approach is also designed to support lower false-alert rates by focusing on predefined safety scenarios and contextual relationships. viAct describes its platform as using scenario-based AI to identify complex micro-actions rather than simply detecting objects, with industry-specific modules trained around real industrial hazards and non-compliances.
What Measurable Impact Did the Deployment Achieve?
The Singapore construction deployment reported a 10× improvement in safety score and 7,000+ working hours saved, alongside on-time project delivery and smoother MOM safety compliance. The platform also created a continuous safety record that could support audits, reporting, and longer-term safety analysis.
Other viAct deployments show how the impact can vary by use case. For example, its Chile mining deployment reported a 70% reduction in proximity risks, while the Doha Metro perimeter-monitoring deployment reported a 70% decrease in manual patrolling and faster emergency response through AI-triggered notifications.
How Does viAct Approach Privacy and Responsible Use of Workplace Video?
Privacy is positioned as part of viAct's privacy-by-design and responsible-AI approach, rather than something added after deployment. The platform supports features such as automated face masking, encryption, and on-premise deployment, allowing organisations to determine how sensitive video data is processed and retained. viAct also describes its system as designed for ethical and compliant AI use in industrial workplaces.
This is particularly important for workplace safety because effective monitoring should not mean unrestricted surveillance. viAct's approach is to use visual data primarily to identify defined safety conditions and risks, while applying controls around access and data handling. Its responsible-AI positioning also includes restricted system access and controlled data transfer, with attention to GDPR and data-privacy requirements.
The broader lesson: the value of computer vision is not measured by how much video an organisation collects, but by how effectively it converts visual data into relevant safety intelligence, measurable improvement, and responsible action.