Decision‐tree‐based human activity classification algorithm using single‐channel foot‐mounted gyroscope

Abstract

Wearable devices that measure and recognise human activity in real time require classification algorithms that are both fast and accurate when implemented on limited hardware. A decision-tree-based method for differentiating between individual walking, running, stair climbing and stair descent strides using a single channel of a foot-mounted gyroscope suitable for implementation on embedded hardware is presented. Temporal features unique to each activity were extracted using an initial subject group (n = 13) and a decision-tree-based classification algorithm was developed using the timing information of these features. A second subject group (n = 10) completed the same activities to provide data for verification of the system. Results indicate that the classifier was able to correctly match each stride to its activity with >90% accuracy. Running and walking strides in particular matched with >99% accuracy. The outcomes demonstrate that a lightweight yet robust classification system is feasible for implementation on embedded hardware for real-time daily monitoring.