Fall Detection Based on Local Peaks and Machine Learning

This research focuses on Fall Detection (FD) using on-wrist wearable devices including tri-axial accelerometers performing FD autonomously. This type of approaches makes use of an event detection stage followed by some pre-processing and a final classification stage. The event detection stage is basically performed using thresholds or a combination of thresholds and finite statemachines. In this research, we extend our previous work and propose an event detection method free of thresholds to tune or adapt to the user that reduces the number of false alarms; we also consider a mixture between the two approaches. Additionally, a set of features is proposed as an alternative to those used in previous research. The classification of the samples is performed using a Deep Learning Neural Network and the experimentation performs a comparison of this research to a published and well-known technique using the UMA Fall, one of the publicly available simulated fall detection data sets. Results show the improvements in the event detection using the new proposals.