Hpe hub banner.png

Health and safety gone rad: Fogsphere brings AI chops to workplace monitoring

Heath and safety gone rad. Source: Fogsphere

Image:

Heath and safety gone rad. Source: Fogsphere

London-based startup Fogsphere uses standard 4K security cameras - the type that can be purchased for £70 - to monitor workplaces and create immediate alerts as soon as dangerous conditions are spotted.

For example, a worker entering a construction site without a hard hat, a machine tool operator without gloves, a forklift truck driving too fast, night workers with no hi-viz clothing, all of these can be picked up with the company's core algorithm, and new specialised use cases can be added as required.

With GenAI incorporated into the cameras, these alerts can be made highly contextual by training the model on specific data. For example a forklift on its side might be having a wheel changed, but if there is a pool of liquid by it and someone lying on the floor, that could be something else entirely, and the event can be immediately raised to an emergency. The question then is how to scale that out across the workspace.

Fogsphere grew out computer vison research at Aberdeen University, Imperial College and UCL. "In 2018 it was all about counting people and cars," said CEO Pasquale Giampa, a former researcher at Imperial. "But no-one was interested in that any more. So I came up with the idea that there was a problem in safety, to use this to save people lives in the workplace."

One challenge is the variety of locations. A hospital in a city centre with reliable broadband is one thing, an energy installation in a desert is quite another. New use cases can be covered by new models trained to spot danger in the context, but connectivity can become a limiting factor. Which is why Fogsphere reached for the term "fog computing" coined by Cisco in 2014, but largely forgotten.

The "fog" is the bit between the cloud and the edge device. "Edge doesn't scale", explained Giampa. "Fog does scale. The fog is basically extending the cloud."

Fogsphere's software is installed at the edge (on the camera), in the fog (for example, in a datacentre in a power plant), and on a public cloud to enable coordination across and between sites.

In remote locations, connectivity may be by private 5G network provided by implementation partner HPE, which rakes care of the hardware side. "We're a software company, not an integrator," said Giampa.

The big advantage the fog computing setup over the pure SaaS favoured by others in this space is that it allows for the use of hi-res camera footage to trigger real-time alerts in challenging outdoor environments without loss of detail.

4K 720p cameras are now standard, but transferring data to the cloud for analysis requires reducing the frame rate to overcome data transfer issues, Giampa explained, often to down from 40 or so to eight frames a second, and that reduces granularity.

"It's physics. You can measure a forklift speed because with that type of frame rate you have 1/8 of a second so you can track up to 2.5 miles an hour, but if you go outdoors, to a power plant, and you want to measure the speed of a car on this very long road you can't. The resolution of the video is not high enough."

Fogsphere's first customer "the largest water desalination and power generation company in the Middle East," was signed up after a demo on the basis this hi-res capability, Giampa claimed.

Surveillance fears

In his native Italy, however, things did not go quite so smoothly, with rollout in a railway setting meeting resistance from the unions. "For a variety of reasons they are very afraid of computer vision in general," said Giampa.

This is understandable in many ways. No-one wants a spy in the sky, but Giampa insisted that in this case the concern is misplaced. Fogsphere's platform is only interested in aggregate health and safety events, helping companies meet health and safety KPIs, as is increasingly required by auditors. It automatically blurs faces, and the ability to track individuals cannot be retrofitted into the system.

"If you're blurring the face, then basically the platform loses a lot of value to somebody that wants to surveil," he said. "We don't care if it's John or Jane that didn't get killed today. It's about neither of them getting killed."