A dual model approach to EOG-based human activity recognition. (August 2018)

• Record Type: Journal Article
• Authors: Lu, Yu; Zhang, Chao; Zhou, Bang-Yan; Gao, Xiang-Ping; Lv, Zhao
• Is Part Of: Biomedical signal processing and control. Volume 45(2018)
• Page Start: 50
• Page End: 57
• Publication Date: 2018-08
• Subjects: Electrooculography (EOG) -- Human activity recognition -- Confidence parameters -- Signals recognition -- Contextual relationship -- Support vector machine; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2018.05.011
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: We propose a dual model to achieve EOG-based human activity recognition. "Activities Relationship" model is established to describe the contextual relationship among different activities. We introduce the confidence parameter to fuse "EOG Signals Recognition" model and "Activities Relationship" model. Experiments show the superiority of the proposed method over related methods. Abstract: Ongoing eyeball activities can be recorded as Electrooculography (EOG) to discover the links between human activities and eye movements. In the present work, we propose a dual model to achieve human activity recognition (HAR) under a specific task background. Specifically, the "EOG Signals Recognition (ESR)" model is used to recognize basic eye movement unit signals collected under different activities; the "Activities Relationship (AR)" model is utilized to describe the contextual relationship among different activities. Furthermore, we introduce a confidence parameter to comprehensively analyze and judge outputs of the above two models. To evaluate the performance of the proposed algorithm, the experiments have been performed under an office scene over 8 subjects. The average recognition accuracy achieves 88.15% according to 3 types of activities (i.e., reading, writing, and resting). Experimental results reveal that the EOG-based dual model HAR algorithm presents an excellent classification performance.