Abstract
Better life of human is a central goal of information technology. To make a useful technology, in sensor network area, activity recognition (AR) is becoming a key feature. Using the AR technology it is now possible to know peoples behaviors like what they do, how they do and when they do etc. In recent years, there have been frequent accidental reports of aged dementia patients, and social cost has been increasing to take care of them. AR can be utilized to take care of these patients. In this paper, we present an efficient method that converts sensor's raw data to readable patterns in order to classify their current activities and then compare these patterns with previously stored patterns to detect several abnormal patterns like wandering which is one of the early symptoms of dementia and so on. In this way, we digitalize human activities and can detect wandering and so can infer dementia through activity pattern matching. Here, we present a novel algorithm about activity digitalization using acceleration sensors as well as a wandering estimation algorithm in order to overcome limitations of existing models to detect/infer dementia.