Matrix Profile based Anomaly Detection in Streaming Gait Data for Fall Prevention

TL;DR

This paper introduces a gait anomaly detection system using the Matrix Profile algorithm, capable of real-time, personalized fall prevention by efficiently identifying abnormal steps in streaming gait data.

Contribution

It presents a novel, efficient, and adaptive gait anomaly detection method based on Matrix Profile, suitable for edge deployment and outperforming neural network baselines.

Findings

Outperforms neural network baseline in anomaly detection accuracy

Operates with low latency suitable for real-time applications

Adapts to individual gait patterns for personalized detection

Abstract

The automatic detection of gait anomalies can lead to systems that can be used for fall detection and prevention. In this paper, we present a gait anomaly detection system based on the Matrix Profile (MP) algorithm. The MP algorithm is exact, parameter free, simple and efficient, making it a perfect candidate for on the edge deployment. We propose a gait anomaly detection system that is able to adapt to an individual's gait pattern and successfully detect anomalous steps with short latency. To evaluate the system we record a small database of enacted anomalous steps. The results show the system outperforms a more complex Neural Network baseline.