Stacking spatial-temporal deep learning on inertial data for human activity recognition


Chew, Yong Shan and Pang, Ying Han and Ooi, Shih Yin and Poh, Quan Wei (2022) Stacking spatial-temporal deep learning on inertial data for human activity recognition. Journal of Engineering Science and Technology., 17 (5). pp. 3235-3253. ISSN 1823-4690

[img] Text
Restricted to Repository staff only

Download (947kB)


Insufficient physical activity has negative effects on quality of life and mental health. Further, physical inactivity is one of the top ten risk factors for mortality. Regular recognition and self-monitoring of physical activity are in the hope to encourage users to stay active. One such application is through intelligent human activity recognition which is usually embedded in ambient assisted living systems. A spatial-temporal deep learning is proposed in this paper for smartphone-based intelligent human physical activity recognition. In this work, a stacking spatial-temporal deep model is devised to extract deep spatial and temporal features of inertial data. In the proposed system, a convolutional architecture is pipelined with Bidirectional Long Short Term Memory to encapsulate the spatial and temporal state dependencies of the motion data. Support Vector Machine is adopted as the classifier to distinguish human activities. Empirical results demonstrate that the proposed system exhibits promising performances on two public datasets (UC Irvine dataset and Wireless Sensor Data Mining database) with 92% and 87% accuracy, respectively.

Item Type: Article
Uncontrolled Keywords: Bidirectional long short term memory, Convolutional neural network, Human activity recognition, Inertial data, Smartphone
Subjects: Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions: Faculty of Information Science and Technology (FIST)
Depositing User: Ms Nurul Iqtiani Ahmad
Date Deposited: 01 Dec 2022 03:16
Last Modified: 01 Dec 2022 03:16


Downloads per month over past year

View ItemEdit (login required)