Towards Audio-Based Emergency Detection for Ambient Assisted Living Applications
Tobias Rothmeier, Stefan Kunze · 2025
In order to enable elderly or people in need of care, to stay living in their homes, even when alone, digital support systems utilizing the Internet of things and advances in embedded computing and artificial intelligence is a promising approach. In this work, a baseline prototype of a wireless acoustic sensor network is presented. It is able to classify environmental sounds including emergency related sounds like falls in real-time. An NVIDIA Jetson Nano as embedded device uses a ShuffleNet V2 model trained in advance for audio classification. The system analyzes the input of three independent microphone nodes based on ESP microcontrollers in parallel and has additional security added to it due to the use of the WireGuard VPN.