Robust recognition of human activities using smartphone sensor data
Ramy Riad Hussein, Jianzhe Lin, Kenneth Michael Madden, Z. Jane Wang · 2017
Human Activity Recognition (HAR) plays a key role in remote patient activity monitoring, particularly for the disabled and elderly. Manually supervising human's activity would be labor and time-consuming. To relieve this manual process, machine learning-based HAR has drawn researchers' attention. In this work, we introduce a robust HAR system capable of identifying six different classes of body movements with high accuracy. The input data is first collected by a smart phone worn on the subject's waist, where the tri-axial accelerometer signals are recorded. Then, the most discriminant features are extracted and further selected from the captured smart phone sensor data, which are further classified by the random forest classifier. We examine the classification performance of the proposed method on a publicly available HAR dataset. The superiority of the proposed method is verified by comparing its performance to those of the state-of-the-art methods. An overall classification accuracy of 98.05% can be achieved.