In food production, sanitisation is not a background task it is a regulatory requirement with direct consequences for product safety. Facilities processing meat, dairy, or ready-to-eat goods must clean and disinfect equipment and production surfaces at defined intervals and according to documented procedures.
The challenge is verification. Across a large production floor and multiple shifts, manual supervision alone rarely provides reliable real-time oversight. An operator cannot simultaneously monitor foam coverage, confirm PPE compliance for every worker, and maintain audit-quality records.
Computer vision and machine learning make this process measurable and automated using existing camera infrastructure without interrupting operations.
Two Phases, One Monitoring System
An AI sanitisation monitoring system typically covers two process stages:
- Water-based cleaning workers spray down equipment and surfaces.
- Foam application cleaning agents are applied to target areas with required coverage thresholds.
Each stage requires different monitoring logic but runs through the same real-time video analytics pipeline.
During water cleaning, the system detects hose activity, tracks liquid application, estimates flow rate in litres per minute, and records which surfaces were cleaned.
During foam application, the focus shifts to surface coverage. The system calculates the percentage of coated area, checks whether thresholds are met, and flags foam appearing outside intended zones, such as floor pooling or overspray.
Coverage zones are highlighted directly on the live feed, creating a continuous spatial record of the cleaning process.
Rather than acting only as anomaly detection, the system continuously documents what happened, where, and when throughout the cleaning cycle.
Worker Safety Without Surveillance
Sanitisation involves exposure to pressurised water, chemicals, and foam compounds, making PPE compliance critical.
The AI layer automatically verifies required protective equipment for each worker individually, including respiratory protection, gloves, footwear, and protective clothing. Violations are linked to a worker’s position within the monitored zone instead of generating generic alerts.
People counting provides real-time occupancy data and supports both zone management and audit reporting.
Worker analysis is performed using pose estimation and body-part segmentation, identifying the head, hands, torso, and legs to confirm PPE presence.
Privacy is preserved throughout the process. Faces are automatically blurred before footage is stored or displayed. Detection relies on body-level signals, posture, movement, and equipment, so privacy measures do not reduce monitoring performance.
When PPE gaps are detected, alerts are generated automatically and recorded in timestamped logs.
Liquid Intelligence Across Surfaces
One of the most demanding technical challenges is distinguishing liquids and measuring their behaviour in real time.
Computer vision models classify water and foam separately across equipment and floor surfaces. This classification determines which measurements and compliance thresholds apply.
Hose detection and segmentation identify both the hose and the liquid stream. Flow rate estimates are generated directly from visual characteristics without requiring additional hardware.
Water and foam coverage zones are displayed as overlays on the video feed, giving operators an immediate understanding of cleaning progress.
During foaming stages, the system calculates exact coverage percentages, verifies completion, and detects foam outside approved areas. Floor pooling and overspray events are logged for both safety and compliance reporting.
The Computer Vision Stack
The system combines custom-trained object detection and segmentation models operating on live camera feeds.
YOLO handles person detection and initial region identification. A dedicated industrial body-part segmentation model supports PPE verification by analysing individual body regions.
Hose and liquid segmentation rely on SAM-class architectures to generate pixel-level masks required for accurate coverage measurement.
Metrics such as flow rate, foam coverage, and surface area are calculated by dedicated computer vision algorithms built specifically for sanitisation monitoring rather than general-purpose AI outputs.
The pipeline operates in real time and writes results directly into a timestamped event system.
The architecture is designed to work with existing surveillance infrastructure. Models are trained on facility-specific data to account for differences in lighting, equipment geometry, and cleaning agent appearance, enabling accurate measurements rather than approximate detections.
Audit Trails and Process Consistency
Operationally, the system generates a structured record of every cleaning cycle: who was present, what PPE was worn, where liquids were applied, achieved coverage levels, and any detected deviations.
The process removes reliance on manual reporting while preserving human accountability. Floor managers, QA teams, and auditors receive a verifiable record of actual activity.
For facilities subject to inspections or operating across multiple shifts and sites, AI-powered video analytics transforms process adherence from policy-based assurance into documented evidence, timestamped, spatially mapped, and generated automatically.
It’s time to work smarter
If you’d like to see how this works in a facility like yours, request a demo.