Behaviour analysis and reminder delivery in an assisted environment

Abstract

Chronic diseases are costly for global health systems and society in general. Many of the associated health problems can be prevented by adherence to healthy lifestyle. This research undertakes the design and evaluation of a behaviour analysis and reminder delivery framework, referred to as iMessenger (intelligent messenger). The research comprises four topics: activity classification; location detection; context management; and context-aware reminders. Feedback can be devised with the intention to promote people to lead a healthier lifestyle. Three algorithms were developed and evaluated for activity classification: a hierarchical algorithm for motion and motionless posture recognition; a fall detection algorithm; and a novel Pick-out algorithm for optimal SVM model selection. The posture classification accuracy achieved using a conventional training set (70.3%) was improved to 85.1% for an optimal training set. Three solutions were used to improve localisation accuracy and robustness based on radio frequency identification (RFID) technology: optimization of RFID reader network deployment; a pre-processing algorithm for handling missing data and a subarea mapping method evaluated using different levels of localisation resolution. The average localisation accuracy based on pre-processed and original data was 77.1% vs. 68.7% for fine-grained and 95.8% vs. 85.4% for coarse-grained subareas. Ontological modelling and reasoning were utilized to integrate heterogeneous contexts, and to infer: whether a user maintains healthy postures; whether a fall has occurred, or whether an abnormal activity has happened and whether people comply with their prescriptive schedule. Algorithms were evaluated using simulated scenarios. The inference results were generated in the form of three types of reminders: unhealthy posture reminder, fall alert, and event inconsistency reminder. iMessenger has the potential to encourage people to adopt a healthier lifestyle for enhanced well-being and can therefore potentially reduce the burden of healthcare. Future work will extend it to support real-time reminder delivery.

Date of Award	May 2011
Original language	English
Sponsors	Vice Chancellor's Research Scholarship (VCRS)
Supervisor	Paul Mc Cullagh (Supervisor), Christopher Nugent (Supervisor), Huiru (Jane) Zheng (Supervisor) & Norman Black (Supervisor)