— July 2nd, 2021
A dangerous fall can happen to anyone, but they are particularly dangerous among the elderly as that demographic might not have effective ways to get help when needed. Rather than having to purchase an expensive device that costs up to $100 per month to use, Nathaniel F. on Hackster wanted to build a project that harnessed the power of embedded machine learning to detect falls and send an alert. His solution involves the Arduino Nano 33 BLE Sense board, which not only has an integrated accelerometer but also contains Bluetooth Low Energy capabilities that lets the processor communicate with the accompanying mobile app.
Nathaniel trained his ML model on the SmartFall dataset, which allows the device to respond to a wide variety of falls and ignore non-harmful movements. Once training was completed, he was able to achieve an accuracy of 95%. The Nano 33 BLE Sense samples accelerometer data at 31.25Hz to match the dataset’s frequency, and it makes a prediction every two seconds. If a fall is detected or the built-in emergency button was pressed, the user has 30 seconds to deactivate the alarm, otherwise it sends a BLE message to the phone which in turn sends an SMS message to an emergency contact containing the current location.
Even though this DIY fall detector works well already, Nathaniel plans on making a custom PCB and extending the battery life for longer use time between charging. You can read more about his design here, and you can view his demonstration video below.
Website: LINK
