Swedish start-up Swanholm AB has developed a smart safety vest for construction workers that utilises TinyML technology to send alerts if a worker falls or has an accident. Swanholm incorporated TinyML technology and AI training from Stockholm based edge AI and TinyML specialist Imagimob AB.

According to Eurostat, the EU statistical agency, across the EU, construction records the highest number of fatal accidents of any occupation, with 21.7% of all fatal workplace accidents being in the construction industry, and the third highest rate of non-fatal accidents, with 12.4%.

The smart safety vest complies with EN 471 class 2 safety standard and has been developed in collaboration with vest manufacturers, embedded engineers, transport and construction companies and end-users. In parallel with these interviews and workshops, the software performing the detection of hazardous falls needed to be developed and trained. Swanholm founder Jonas Svanholm brought in Imagimob in to research the needs end-users,  which led to the development of the TinyML employed in the smart vest.

TinyML is AI that can run on embedded and wearable systems. Typically, TinyML models are trained on large conventional ML platforms, whether on the cloud or on-site, and then the trained production AI system is minified for embedded systems and transferred to an embedded or wearable chip.

One of the best known available datasets for fall detection is the UR fall detection set developed by a team at University of Rzeszow in Poland. The accelerometer used in their study, however, was placed at the waist and not at the neck as with the Connected Safety Vest’s sensor mount point, rendering such a third-party dataset less accurate for this purpose, but still useful to some degree, according to Imagimob. Differences in sensor type employed meant that more training data was required in the AI model.

The sensor used in the vest is the Bosch BMI-160, which carries both 3-axis accelerometer and 3-axis gyroscope in a small form factor. It handles acceleration up to ±16g and its power consumption is 180 μA for the accelerometer in full operation and 3µA in suspended mode, making it a good candidate for embedded applications when powered by batteries.

After data collection and labelling, the model is generated and trained followed by verification using testing data. The last step is deployment into a executable binary library.

Not all falls result in injury, so the device is equipped with a timer which starts right when a fall is detected, allowing for the user to reset the vest during a grace time, during which a set of warning beeps are heard. Only after 30 seconds is an alarm issued, sending notifications to peers, management and/or safety staff. If there’s an actual accident the user does not need to do anything as the alarm will execute automatically.

The connected vest also is equipped with a push-button switch on the front, to raise attention should an operator become stuck, wedged or otherwise pinned in a work situation without being able to call someone using their phone. There might not be a fall at all, and this makes up a secondary way of calling for assistance to complement the AI detection algorithm.

Firmware on the vest can be modified and updated securely over the air using a smartphone. This allows for new tinyML models and applications to be updated as the fall detection algorithms evolve without having to hand them in for an upgrade.