Swanholm Tech's Connected Safety Vest Is a Wearable TinyML Lifesaver

Swanholm Tech, founded by former Scania employee Jonas Svanholm, working with edge AI and tinyML specialist Imagimob, has come up with something designed to make truck drivers safer: a smart safety vest, capable of detecting a fall and triggering a call for help.

The types of trucks used for large-scale logistics are absolute beasts, with high cabs and even higher loads. For a driver or loader who falls, that's a potentially lethal situation — with one wheel loader operator at an unnamed company claimed by Swanholm Tech to have fallen around six and a half feet after experiencing a dizzy spell, then lying behind the wheel for a full hour prior to being discovered and assisted.

To solve the problem, Svanholm founded Swanholm Tech and set about developing a wearable machine learning system, which could detect falls and call for help automatically — using on-device tinyML technology from Imagimob. Based on a bright-yellow high-visibility vest, the Connected Safety Vest uses a Bosch Sensortec BMI160 inertial measurement unit (IMU), chosen for it small size and power draw.

To train the machine learning model, a combination of the KFall fall-detection dataset and hands-on training — involving Swanholm staff literally jumping, rolling, collapsing, and at one point judo-rolling in a variety of ways before a crash-test dummy was brought in for the more dangerous stunts. A broad dataset for training and test was vital, not only to accurately detect falls but to avoid false positives — workers who can't trust their vests not to cry wolf won't wear them for long.

When a fall is detected, the vest starts a countdown timer. If the wearer doesn't disable the timer before it expires, the alarm is raised and staff notified to attend. For emergencies that don't trigger the fall alarm, such as becoming stuck within a load, a physical push-button on the vest allows the alert to be triggered manually.

More information on the Connected Safety Vest is available on the Swanholm Tech website.