This GIGA R1 WiFi-powered wearable detects falls utilizing a Transformer mannequin

For these aged 65 and over, falls might be some of the severe well being considerations they face both because of decrease mobility or reducing general coordination. Recognizing this situation, Naveen Kumar got down to produce a wearable fall-detecting system that goals to extend the velocity at which this happens by using a Transformer-based mannequin somewhat than a extra conventional recurrent neural community (RNN) mannequin.

As a result of this mission wanted to be each quick and eat solely small quantities of present, Kumar went with the brand new Arduino GIGA R1 WiFi because of its STM32H74XI dual-core Arm CPU, onboard WiFi/Bluetooth®, and skill to interface with all kinds of sensors. After connecting an ADXL345 three-axis accelerometer, he realized that accumulating many hours of samples by hand could be far too time consuming, so as an alternative, he downloaded the SisFall dataset, ran a Python script to parse the pattern information into an Edge Impulse-compatible format, after which uploaded the ensuing JSON information right into a new mission. As soon as accomplished, he used the API to separate every pattern into four-second segments after which used the Keras block edit characteristic to construct a reduced-sized Transformer mannequin.

The consequence after coaching was a 202KB giant mannequin that would precisely decide if a fall occurred 96% of the time. Deployment was then so simple as utilizing the Arduino library characteristic inside a sketch to run an inference and show the consequence through an LED, although future iterations may leverage the GIGA R1 WiFi’s connectivity to ship out alert notifications if an accident is detected. Extra info might be discovered right here in Kumar’s write-up.