Researchers have developed a neural network for recognizing human activity based on data from wearable devices. The project is capable of tracking rapid changes in movement and analyzing their sequence. Thanks to this, the system can distinguish walking from running and detect significant changes in a user's activity.
The model contains only 50,000–62,000 parameters and can run on smartwatches, fitness trackers, and medical devices. Across three research datasets, its accuracy reached 98.91%.
Such a system can quickly detect falls and other potentially dangerous movements and immediately trigger the necessary action on the device without sending data to the cloud. This could provide new opportunities for analyzing human activity directly on wearable devices.