Activation-based recurrent learning method for wearable sensor data processing in human activity recognition

摘要

Human activity recognition using wearable sensors is a universal application for providing additional real-time support in AAL and medical healthcare systems. Input signals and sensor data are analyzed for the features, followed by correlation analysis to achieve better recognition accuracy. In this article, activation-based recurrent learning for human activity detection is introduced. This method is based on recurrent neural networks that pre-classifies the features and extract them for time and rate based on the input signals. Following the pre-classification, the repetitive learning estimates the features for the correlation and activities to reduce false-negative errors in processing. This recurrent process is validated using a conditional activation function and rectified linear unit for joint and concurrent layer processing. The combined processing feature of this proposed method helps to improve the accuracy by 92% and recall by 94.5% by reducing the computation time. The performance of the proposed method is analyzed for six activities, varying iterates, and at time intervals.

关键词