Human CCTV scans ‘unreliable’

Surveillance of abattoir CCTV is not reliable, according to an Australian-based pioneer of  camera-based AI systems.

“Humans are not good at watching videos. We humans are subjective and inconsistent,” said Jamila Gordon, founder and CEO of Lumachain.  “It’s not real-time. You are watching a video after the event. It’s night and day, It’s 24/7. It takes time. And it’s expensive.

“If the AI detects an occurrence there’s an immediate alert sent to the supervisor and action is taken. So things never get too bad. Humans? We drift.”

The company she founded in 2018 has grown to include a customer list that reads like a food industry Who’s Who.

JBS uses Lumachain technology to track every shipment to multi-national US food chain Chipotle Mexican Grill.

Jamila said: “Chipotle invested in us because of that traceability. Not only are we tracing the products – in every factory we have computer-vision AI for safety, quality and record every condition about that product.

“McDonalds fell in love with us and referred us to their suppliers like National Beef in the US which also works with us. They also introduced us to Coca-Cola.”

HOW DOES AI WORK?

Abattoir cameras working with day and night-vision are coupled with AI to monitor animal welfare and intrusion protection 24/7.

Modules within the platform include electric prodding, defective stunning, potential egregious acts, pen density, animal counting and – under construction – bleeding rail sensibility.

“AI learns from data.” Jamila said: “We taught the AI what good looks like. And we said ‘if you see anything that’s not good, bring it to the attention of a human’. It was slow. The first project took 18 months. Now, because we have had so many practicals and we’ve collected enough data, it takes four weeks.”

A Colorado State University study concluded that the AI response is now is similar to a certified animal welfare auditor with 10-years-plus experience.

The same technology can be used to monitor the performance of boning room staff or for foreign object detection, she added.