AI on the food safety and worker hygiene job

The work has paved the way for future intelligent monitoring systems within the red meat processing industry and shown that AI could help transform video monitoring from passive observation into operational decision support.

Conducted by Advance Analytics Australia, and funded by AMPC, the eight-month research project has shown artificial intelligence can play a valuable role in monitoring hygiene and safety compliance in red meat processing facilities.

AMPC Program Manager Markets and Integrity Ann McDonald said while processors and workers well understood how essential hygiene and safety compliance was, monitoring it continuously and consistently in such a dynamic environment and supporting workers to improve, if necessary, was challenging.

The project demonstrated AI-assisted monitoring could provide structured visibility of personnel movement, personal protection equipment use, sanitation behaviour and handwashing activity.

Computer vision systems could provide value beyond simple detection tasks when they are designed around operational workflows, behavioural context and reviewable reporting, AMPC Program Manager Markets and Integrity Ann McDonald said.

"This technology could potentially improve worker hygiene and reduce resourcing requirements for plants to monitor," she said.

The project piloted a modular AI-powered compliance monitoring system called SafePassAI, designed by Advance Analytics Australia, at a red meat processing site.

Advance Analytics Australia co-founder Fred Asgari said the system leverages Axis cameras with system-on-chip capabilities that run custom AI applications directly on the device. These on-camera applications analyse video footage in real time using trained machine learning models to detect compliance behaviours such as handwashing and boot sanitisation.

Rather than streaming full video to a central server, the cameras generate structured metadata—compact text-based records containing compliance outcomes and activity timestamps, which are transmitted to a central high-performance server. This architecture dramatically reduces network load while improving processing efficiency and real-time responsiveness.

"A key innovation behind SafePassAI is its ability to track personnel across multiple cameras without depending on visual appearance or clothing recognition," Mr Asgari said.

"In processing facilities where workers wear near-identical PPE, traditional tracking approaches become significantly less effective.

"SafePassAI instead uses sophisticated movement-based and behavioural tracking techniques that analyses how individuals move through space and interact with hygiene stations in sequence. This allows the system to maintain workflow continuity for each individual and simultaneously validate hygiene compliance for high volume of  personnel in real time.”

The server aggregates metadata from all cameras and enables advanced validation and reporting. A web-based dashboard allows operators to review compliance status, trigger alerts and analyse historical trends.

"A key conclusion is that practical compliance monitoring requires far more than isolated AI detections; it requires an intelligent system capable of automatically improving staff processes and driving real-time corrective action without the constant need for supervisor intervention,” Mr Asgari said.

"The project showed that meaningful operational insight is achieved when visual observations are combined with tracking, behavioural interpretation, session-level aggregation, reporting and human review.

"The project also demonstrated that behavioural AI can operate practically in complex industrial environments when designed around real-world constraints. These constraints include visual obstruction, personnel overlap, changing environmental conditions and variation in how procedures are performed."

IMAGE CAPTION: SafePassAI hand detection output used to support handwashing activity analysis.