Research on Human Activity Recognition Method Based on Wearable Sensors

Wearable sensors have become essential platforms for research on human activity recognition (HAR) due to their advantages such as compact size, low power consumption, and non-invasive nature, ensuring continuous and privacy-conscious data acquisition. In this paper, a HAR architecture using Machine Learning (ML) technique based on data collected from wearable sensors is proposed to perform high performance for accurate recognition of human activities in real-life scenarios. To address the challenge of accurately distinguishing similar forms of daily activities, a feature library consisting of 55 feature functions has been constructed, and the Maximum Relevance Minimum Redundancy (mRMR) algorithm is employed to select the most informative and relevant features for activity classification. Experimental results indicate that the combination of data from multiple sensors and the dynamic selection of features significantly improve the performance of the HAR system, as they provide a more comprehensive and diverse set of information. The numerical results show that the human activity recognition framework proposed in this paper can achieve an accuracy of 98.53% on the self-collecting dataset.