Leak & Hazard Detection is a collection of AI-based video analytics technologies designed to identify selected visual conditions associated with fluid leakage, gas releases, and chemical exposure.
Industrial facilities, manufacturing plants, laboratories, utility installations, and storage environments contain equipment and materials that may develop leaks or hazardous conditions during normal operations.
Traditional surveillance cameras record these environments, but identifying a small leak or subtle visual change can require continuous human observation.
AI video analytics adds an automated monitoring capability by continuously analyzing surveillance images for predefined visual patterns.
The solution includes four applications:
Oil Leak Detection
Water Leak Detection
Gas Leak Detection
Test Paper Color Change by Chemical
Although all four applications use video analytics, their underlying recognition methods differ.
Oil and water detection focus on visible liquid appearance and changes in the monitored scene.
Gas Leak Detection focuses on visible release phenomena that can be recognized by the deployed imaging system.
Test Paper Color Change by Chemical focuses on changes in the appearance of a physical chemical indicator.
Together, these applications provide complementary approaches to monitoring visually observable hazards.
Common AI Technology Architecture
The four applications can be understood through a shared processing framework:
Video acquisition → monitoring-region analysis → visual condition recognition → temporal evaluation → event generation
The exact recognition logic depends on the application.
Video Acquisition
Surveillance cameras continuously capture images of designated equipment, surfaces, or monitoring locations.
Typical targets include:
Pipes and valves
Pumps and compressors
Machinery
Storage containers
Industrial floors
Utility equipment
Chemical test indicators
Other locations where leaks may develop
A stable camera view is particularly valuable because many leak-related conditions are identified through changes in the appearance of a known location.
Region-of-Interest Analysis
The system can focus analysis on a configured monitoring zone.
For example, a camera viewing an entire mechanical room may only need to analyze the floor beneath a particular pump.
Restricting analysis to the relevant area helps reduce interference from unrelated movement elsewhere in the image.
Visual Feature Recognition
The AI evaluates visual characteristics associated with the configured hazard.
Depending on the application, these may include:
Liquid appearance
Surface texture
Reflective patterns
Changes in color
Expanding wet regions
Plume-like formations
Irregular motion
Changes in a test indicator's color
No single characteristic is sufficient for every application.
The analytics therefore relies on visual patterns represented by the model and the evidence available in the camera image.
Temporal Analysis
Leaks and chemical changes often develop over time.
A surface may gradually become wet, a liquid region may expand, or a test paper may change color after exposure.
Analyzing consecutive frames can help distinguish these developments from brief shadows, reflections, or passing objects.
Event Generation
When the configured visual criteria are satisfied, the system generates an event.
Depending on the deployment, this event can be used for notifications, recording, video bookmarking, and operational review.
- Oil Leak Detection
Technology Principle
Oil Leak Detection identifies visual patterns associated with oil appearing or accumulating within a monitored area.
The technology may combine AI-based liquid recognition, surface appearance analysis, spatial evaluation, and temporal scene-change analysis.
A simplified sequence is:
Normal equipment area → oil-like liquid appears → visual characteristics analyzed → leak condition confirmed → Oil Leak Detection event
Visual Characteristics of Oil
Oil can have several recognizable visual properties, depending on its type and the surface on which it appears.
These may include:
Dark or translucent liquid regions
Glossy or reflective surfaces
Irregular puddle boundaries
Gradually expanding liquid areas
Streaks or flow patterns
Changes in the appearance of the floor or equipment
Some oils may appear dark, while others are relatively transparent.
This variation makes it important to evaluate multiple visual characteristics rather than relying exclusively on color.
Surface Appearance Analysis
A dry industrial floor typically has a relatively consistent visual appearance.
When oil accumulates, the surface may become darker, more reflective, or visually different from its previous condition.
AI analytics can identify these changes when they correspond to learned oil-leak patterns.
Conceptually:
Dry surface → localized appearance change → oil-like region recognized → event
Liquid Accumulation Analysis
A leak may begin as a small spot and gradually spread.
Temporal analysis can evaluate whether the suspected liquid region persists or expands.
For example:
T1: Surface appears normal
T2: Small liquid region appears
T3: Liquid region expands
T4: Oil-leak condition recognized
This progression can provide additional evidence that a developing leak is occurring.
However, not every leak expands visibly. Some remain localized or occur intermittently.
Distinguishing Oil from Similar Conditions
Potentially confusing visual conditions include:
Water
Cleaning fluids
Shadows
Reflections
Dark floor markings
Surface stains
Wet patches
Changes in illumination
AI models can learn visual differences between oil-like conditions and common background appearances, but classification remains dependent on the quality and characteristics of the available imagery.
A regular camera should not be assumed to identify the precise chemical composition of a liquid.
Camera Placement
Oil Leak Detection is particularly suitable for fixed monitoring locations such as:
Beneath hydraulic equipment
Around pumps
Near pipe connections
Under machinery
Around storage containers
Along designated equipment floors
The camera should provide a clear view of the surface where leaked oil is expected to appear.
- Water Leak Detection
Technology Principle
Water Leak Detection identifies visual patterns associated with water appearing, flowing, or accumulating within a monitored area.
The technology may use AI-based liquid recognition, surface-change analysis, spatial evaluation, and temporal monitoring.
The basic process is:
Normal dry area → water-like condition appears → visual changes analyzed → Water Leak Detection event
Visual Characteristics of Water
Water can be difficult to recognize because it is often transparent.
Its appearance depends strongly on the surrounding surface and lighting.
Possible visual indicators include:
Reflective wet surfaces
Puddles
Flowing liquid
Expanding wet regions
Surface darkening
Ripples
Changes in reflected light
The AI evaluates combinations of these visual characteristics according to the trained model.
Scene-Change Analysis
A fixed camera can provide a consistent reference for the monitored surface.
When water begins accumulating, the appearance of the scene may change.
For example:
Dry floor → wet region appears → region persists or expands → Water Leak Detection event
This is particularly useful in equipment rooms and other locations where the normal scene remains relatively stable.
Flow and Motion Analysis
Some water leaks produce visible movement.
Examples include:
Water flowing along a floor
Dripping from a pipe
Water spreading from equipment
Small streams developing near a connection
Where the movement is sufficiently visible, temporal analysis can provide additional evidence of leakage.
However, small droplets or transparent streams may be difficult to recognize, particularly at long camera distances.
Distinguishing Water from Other Visual Conditions
Potential false detections may arise from:
Reflections
Shadows
Floor cleaning
Condensation
Other liquids
Changes in lighting
Wet surfaces caused by normal operations
A wet surface does not necessarily indicate a pipe or equipment failure.
The AI identifies a visual condition that may correspond to water leakage; personnel determine the actual source and significance.
Typical Monitoring Locations
Water Leak Detection can be deployed in:
Mechanical rooms
Utility areas
Production facilities
Warehouses
Equipment rooms
Water-treatment facilities
Pump stations
Pipe and valve installations
The most effective camera position depends on where water is expected to become visible.
- Gas Leak Detection
Technology Principle
Gas Leak Detection identifies visually observable conditions associated with gas or vapor releases.
This application requires an important technical distinction:
A conventional visible-light camera cannot directly recognize every gas leak.
Many gases are transparent and do not produce visible changes in ordinary video.
AI-based Gas Leak Detection using regular surveillance imagery therefore depends on a release producing a recognizable visual phenomenon, such as a visible vapor cloud, mist, or plume.
The simplified process is:
Normal equipment area → visible release appears → plume characteristics analyzed → temporal behavior evaluated → Gas Leak Detection event
Visible Gas-Release Indicators
Depending on the substance and environmental conditions, a release may produce:
Visible vapor
Mist
Condensation clouds
Plume-like formations
Expanding airborne regions
Irregular drifting movement
These visual phenomena may provide evidence of a potential release.
However, a visible plume is not necessarily composed of the gas itself. It may result from condensation, entrained droplets, or other physical effects associated with the release.
Plume Appearance Analysis
A visible release may exhibit characteristics such as:
Irregular boundaries
Semi-transparent appearance
Progressive expansion
Continuous deformation
Movement influenced by airflow
Emergence from a fixed location
The AI can evaluate whether the observed region resembles a release pattern represented by the model.
Temporal Motion Analysis
Motion can help distinguish a plume from static objects.
For example:
T1: Equipment area appears normal
T2: Plume-like region emerges
T3: Region expands or drifts
T4: Release-like behavior persists
T5: Gas Leak Detection event
The AI may evaluate movement direction, shape changes, persistence, and apparent source location.
Distinguishing Gas Releases from Similar Phenomena
Several ordinary environmental conditions may resemble a visible gas release.
Examples include:
Steam
Fog
Dust
Smoke
Vehicle exhaust
Water spray
Industrial process emissions
Condensation
These phenomena can create similar visual patterns.
Site-specific camera positioning, detection zones, and model suitability are therefore especially important.
Invisible Gas Limitations
Many hazardous gases are not visible in standard surveillance video.
Examples can include methane, carbon monoxide, and other gases under ordinary conditions.
If the gas does not create a visible effect, a regular camera cannot reliably determine that the gas is present simply by analyzing the scene.
Specialized optical gas imaging equipment or dedicated gas sensors may be appropriate for applications requiring direct gas detection.
Such equipment operates according to different sensing principles and should not be confused with ordinary visible-light AI analytics.
Gas Identity and Concentration
Even when a visible plume is recognized, ordinary video analytics cannot reliably establish:
Gas identity
Gas concentration
Toxicity
Flammability
Oxygen displacement risk
Exposure level
Whether regulatory limits have been exceeded
These properties require appropriate instrumentation.
Gas Leak Detection through regular video should therefore be treated as supplementary visual monitoring rather than a substitute for required gas-safety instrumentation.
- Test Paper Color Change by Chemical
Technology Principle
Test Paper Color Change by Chemical combines a physical chemical indicator with AI-based visual monitoring.
Certain test papers or indicator materials change color when exposed to a particular chemical or environmental condition.
The camera continuously observes the indicator, while AI analytics recognizes the configured color-state change.
The basic process is:
Normal indicator color → chemical reaction occurs → indicator changes color → AI recognizes configured change → event generated
Separation of Chemical Sensing and Visual Recognition
This application contains two distinct functional stages.
Physical chemical indicator
The test paper reacts to the relevant substance or condition.
AI video analytics
The camera and AI identify the resulting visible color change.
The AI does not independently perform chemical analysis.
Its role is to automate observation of the indicator's visual state.
Reference-Color Monitoring
A test indicator may have a known normal appearance.
The AI can evaluate the monitored region against the expected visual state.
For example:
Reference color recognized → normal condition
Configured reaction color recognized → chemical-indicator event
The precise recognition method depends on the model and the indicator being monitored.
Color Analysis
The system may evaluate characteristics such as:
Hue
Saturation
Brightness
Color distribution
Changes across the indicator surface
Persistence of the changed state
The relevant visual criteria depend on the chemical test medium.
Some indicators exhibit a strong color transition, while others change gradually or develop a localized reaction pattern.
Temporal Confirmation
Color changes may develop over several seconds or minutes.
Temporal analysis can help determine whether the observed change is persistent rather than a temporary lighting effect.
Conceptually:
Normal color → partial change → reaction color develops → configured condition confirmed → event
This is especially useful when the indicator changes gradually.
Lighting and Color Accuracy
Consistent lighting is essential for reliable color-based analytics.
The apparent color of a test paper can change due to:
Illumination intensity
Light-source color temperature
Shadows
Reflections
Camera white balance
Exposure
Image compression
Aging or discoloration of the indicator
A stable camera and controlled illumination can improve recognition reliability.
Camera Placement
The indicator should occupy sufficient pixels within the image for the relevant color change to be clearly observed.
Important considerations include:
Fixed camera position
Consistent viewing distance
Suitable image resolution
Stable illumination
Minimal reflections
Clear visibility of the entire indicator
Protection from unnecessary obstruction
Chemical Interpretation
The meaning of the color change depends on the particular test paper.
Different indicators may react to different substances or conditions.
The AI should therefore be configured according to the actual indicator and its specified interpretation.
The system cannot automatically identify arbitrary chemicals simply because a paper changes color.
Where concentration measurement or quantitative chemical analysis is required, the indicator's validated measurement characteristics and appropriate instrumentation remain necessary.
Comparison of the Four Technologies
Although all four applications use video analytics, they analyze different visual evidence.
Application
Primary Visual Evidence
Analytical Focus
Oil Leak Detection
Oil-like liquid appearance
Surface change, liquid recognition, accumulation
Water Leak Detection
Water-like appearance or flow
Wet-region recognition, scene change, movement
Gas Leak Detection
Visible release plume
Appearance, expansion, motion, temporal behavior
Test Paper Color Change by Chemical
Indicator color transition
Color-state recognition, reference comparison, persistence
This distinction is important when selecting the appropriate application.
A model designed to recognize water accumulation is not automatically suitable for detecting gas releases or chemical-indicator changes.
Each application requires recognition logic matched to its intended visual target.
Fixed Monitoring Zones
Leak & Hazard Detection is particularly suitable for known risk locations.
Examples include:
Floor beneath a pump
Pipe connection
Valve assembly
Storage tank outlet
Equipment base
Chemical indicator
Designated containment area
A fixed monitoring zone allows the AI to concentrate on the part of the scene where the condition is expected to appear.
This can help reduce irrelevant visual activity and improve the consistency of monitoring.
Scene Stability and Background Changes
Many leak applications benefit from a relatively stable scene.
If the camera remains fixed, changes in the monitored area can provide useful analytical evidence.
However, industrial environments are often dynamic.
Workers may move equipment, clean floors, reposition containers, or change the layout of a work area.
These activities can affect the appearance of the monitored region.
Scene-change analysis should therefore be combined with appropriate AI recognition rather than assuming that every visual change indicates a leak.
Temporal Analysis and Event Confirmation
Temporal analysis is useful across all four applications.
For oil and water, the AI can observe whether a liquid-like region persists or spreads.
For visible gas releases, the AI can evaluate plume development and motion.
For chemical test paper, the AI can confirm that the expected color transition has occurred.
The general logic is:
Potential condition recognized → additional frames analyzed → visual evidence confirmed → event generated
The required observation period may differ by application.
A rapidly developing visible release may require different timing from a slowly changing chemical indicator.
AI Confidence and Detection Parameters
AI-based recognition is probabilistic.
Detection confidence depends on the available visual evidence and how well the observed condition corresponds to patterns represented by the model.
Relevant factors include:
Image resolution
Camera distance
Lighting
Surface appearance
Object or leak size
Occlusion
Environmental conditions
Video compression
Model training characteristics
Where supported, confidence thresholds and temporal parameters can be configured to balance sensitivity and resistance to false events.
A more sensitive configuration may recognize subtle conditions earlier but may also generate additional alerts from visually similar phenomena.
A stricter configuration may reduce false events but require more obvious evidence.
Camera Placement and Image Requirements
Camera placement should be designed around the visual condition being monitored.
Oil Leak Detection
The camera should clearly observe the surface where oil is expected to appear or accumulate.
Water Leak Detection
The camera should provide sufficient visibility of relevant floors, equipment connections, or drainage areas.
Gas Leak Detection
The camera should have a clear view of locations where a visually observable release may emerge and develop.
Test Paper Color Change by Chemical
The camera should provide a close, stable, consistently illuminated view of the indicator.
Because these applications have different imaging requirements, a single camera angle may not be suitable for every detection task.
Indoor and Outdoor Operation
Leak & Hazard Detection can be applied in both indoor and outdoor environments when the monitored condition remains sufficiently visible.
Indoor locations may offer more consistent illumination and background conditions.
Outdoor locations can introduce additional challenges such as:
Rain
Wind
Fog
Dust
Sunlight variation
Shadows
Reflections
Water already present on the ground
Changing ambient conditions
These factors are particularly important for visual liquid and plume recognition.
For example, rainwater may create visual conditions resembling a water leak, while wind can rapidly disperse a visible vapor plume.
Deployment testing should reflect the actual operating environment.
Multiple Monitoring Locations
Large facilities may require monitoring at numerous potential leak points.
Different cameras can be assigned to different equipment or hazard locations.
For example:
Camera A monitors hydraulic machinery for oil leakage.
Camera B monitors a utility room for water accumulation.
Camera C monitors a process connection for visible release phenomena.
Camera D monitors a chemical test indicator.
Events from these applications can contribute to a common surveillance monitoring workflow.
Event Generation and Surveillance Integration
When the configured detection criteria are satisfied, the system can generate an event within the ACTi video surveillance environment.
Depending on system configuration, the event may support:
Real-time operator alerts
Maintenance notifications
Camera highlighting
Event-associated recording
Video bookmarking
Searchable event histories
Remote verification
Operational review
Integration with other configured workflows
This allows personnel to focus on detected exceptions rather than continuously observing every monitored location.
Visual Verification
One advantage of camera-based hazard monitoring is that the event can be accompanied by video of the relevant area.
For example, personnel may be able to inspect:
Where liquid appears to be accumulating
Whether the suspected leak is expanding
The apparent location of a visible release
Changes in the surrounding environment
The current appearance of a chemical indicator
Visual verification provides useful operational context.
However, video should not be used as the sole basis for approaching a potentially hazardous release. Personnel should follow established safety procedures, including remote assessment and appropriate instrumentation where necessary.
Real-Time Monitoring
The four applications can continuously analyze surveillance video and generate events when sufficient evidence is available.
The common processing sequence is:
Camera captures scene → AI analyzes configured monitoring area → potential condition recognized → confirmation criteria satisfied → event generated
Actual detection latency depends on:
Visual characteristics of the condition
Video frame rate
Processing resources
Network conditions
Detection thresholds
Temporal confirmation settings
Camera image quality
The objective is to provide timely awareness while reducing unnecessary alerts.
Technical Benefits
Leak & Hazard Detection provides several technical advantages:
Four specialized AI applications: Different visual hazard conditions can be monitored within one surveillance environment.
Automated liquid recognition: Oil-like and water-like conditions can be identified in designated areas.
Visual release monitoring: Observable plume-like gas-release conditions can be recognized.
Chemical-indicator monitoring: Color-changing test paper can be observed automatically.
Temporal analysis: Developing conditions can be evaluated across consecutive frames.
Configurable monitoring zones: Analytics can focus on known risk locations.
Continuous observation: Selected areas can be monitored throughout camera operating hours.
Real-time event generation: Potential hazards can be brought to personnel's attention.
Video verification: Operators can inspect the associated scene.
Searchable event records: Detected conditions can be reviewed later.
Flexible deployment: Different analytics can be assigned to different cameras and locations.
Performance Considerations
Leak & Hazard Detection should not be expected to recognize every leak or chemical hazard under all circumstances.
The system depends on the relevant condition producing sufficient visual evidence.
Potential limitations include:
Leaks hidden behind equipment
Liquids too small to distinguish
Transparent liquids with little visual contrast
Poor lighting
Strong reflections
Moving shadows
Low-resolution video
Heavy occlusion
Similar-looking substances
Invisible gas releases
Plumes confused with steam, smoke, or dust
Chemical indicators too small to evaluate
Changes in camera color reproduction
A regular surveillance camera cannot reliably determine chemical composition, toxicity, gas concentration, or leak rate from visual appearance alone.
The capabilities of each application should therefore be evaluated against the actual monitoring objective.
Operational and Safety Considerations
Leak & Hazard Detection is an AI-assisted visual monitoring technology, not a universal chemical or environmental measurement system.
An AI event indicates that a visual condition corresponding to configured criteria has been recognized.
The event should prompt appropriate review and response according to established site procedures.
For hazardous gases, invisible releases, chemical exposure limits, or conditions requiring certified detection, dedicated sensing instruments and safety systems remain necessary.
Chemical test-paper applications depend on the accuracy, suitability, and maintenance of the physical indicator.
Similarly, visual recognition of an oil-like or water-like region does not establish the exact substance or its hazardous properties.
AI video analytics should complement physical inspections, maintenance programs, environmental monitoring instruments, and required safety systems.
Conclusion
Leak & Hazard Detection combines four specialized AI video analytics applications to provide continuous visual monitoring of oil leakage, water leakage, visible gas-release phenomena, and chemical-indicator color changes.
Oil Leak Detection and Water Leak Detection analyze liquid appearance, surface changes, and accumulation patterns.
Gas Leak Detection evaluates visually observable plume-like releases, while Test Paper Color Change by Chemical uses AI to recognize the visual response of a physical chemical indicator.
Although these applications use different analytical methods, they share a common framework of camera-based observation, visual condition recognition, temporal evaluation, and automated event generation.
With suitable camera positioning, appropriate AI models, clearly defined monitoring areas, consistent imaging conditions, and site-specific testing, ACTi Leak & Hazard Detection can provide an additional layer of operational awareness—helping organizations recognize potential problems earlier, verify visible conditions, and direct personnel toward situations that require investigation.
The technology is most effective when integrated with established maintenance procedures, appropriate physical sensors, and required industrial safety systems.
