A single tri-axial accelerometer-based real-time personal life log system capable of human activity recognition and exercise information generation

• Published in: Personal and ubiquitous computing Vol. 15; no. 8; pp. 887 - 898
• Main Authors: Lee, Myong-Woo, Khan, Adil Mehmood, Kim, Tae-Seong
• Format: Journal Article
• Language: English
• Published: London Springer-Verlag 01.12.2011; Springer Nature B.V
• Subjects: Accelerometers; Bioenergetics; Computer Science; Exercise; Feature recognition; Human motion; Information technology; Mobile Computing; Original Article; Personal Computing; Real time; Recognition; Recording; User Interfaces and Human Computer Interaction; Walking; Activity recognition; Accelerometer; Exercise information; Personal life log
• ISSN: 1617-4909; 1617-4917
• DOI: 10.1007/s00779-011-0403-3

Recording a personal life log (PLL) of daily activities in a ubiquitous environment is an emerging application of information technology. In this work, we present a single tri-axial accelerometer-based PLL system capable of human activity recognition and exercise information generation. Our PLL system exhibits two main functions: activity recognition and exercise information generation. For activity recognition, the system first recognizes a state of daily activities based on the statistical and spectral features of the accelerometer signals. An activity within the recognized state is then recognized using a set of augmented features, including autoregressive coefficients, signal magnitude area, and tilt angle, via linear discriminant analysis and hierarchical artificial neural networks. Upon the recognition of each activity, the system further estimates exercise information that includes energy expenditure based on metabolic equivalents, stride length, step count, walking distance, and walking speed. Our PLL system operates in real-time, and the life log information it generates is archived in a daily log database. We have validated our PLL system for six daily activities (i.e., lying, standing, walking, going-upstairs, going-downstairs, and driving) via subject-independent and subject-dependent recognition on a total of twenty subjects, achieving an average recognition accuracy of 94.43 and 96.61%, respectively. Our results demonstrate the feasibility of a portable real-time PLL system that could be used for u-lifecare and u-healthcare services in the near future.