How tri-axial sensors influenced the location-based heterogeneous activities recognition rates: an exploratory analysis

Abstract

Due to the successful enhancement of context-aware applications in health caring, surveillance, security, sports, behavior analysis, and more, the Human Activities Recognition (HAR) has been noted as an emerging research domain. In this paper, we proposed a deep learning-based novel ConvLSTM-HHAR (Convolutional Long short-term Memory-Heterogeneous Human Activities Recognition) model for the recognition of heterogeneous human activities in indoor and outdoor environments. For automatically extracting efficient features from raw sensor data and identifying activities, the proposed model utilizes the Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM), respectively. We have used the publicly availabel KU-HAR dataset as an experimental dataset which includes a total of eighteen activities (categorized as indoor and outdoor) of 90 subjects collected using a single tri-axial accelerometer and gyroscope sensor. The proposed model has achieved an average accuracy of 99.98%, F1-score of 93.34%, precision of 88.06%, and recall of 100.00% for indoor, outdoor, and indoor + outdoor activities.