Disorientation detection by mining GPS trajectories for cognitively-impaired elders

The aging population is a major concern of human society in the 21st century. Elders who suffer physical or cognitive impairments often have difficulties in navigational tasks and remembering landmarks. As such, disorientation (i.e., getting lost) becomes common for people with dementia in unfamiliar or even in familiar environments. In order to provide appropriate real-time assistive services to these elders, such as reminders and alerts, we propose a disorientation detection method that detects outliers in one's GPS trajectories. In particular, we first model an individual's movement trajectories as a graph based on her historical GPS traces. We then identify outlying trajectories that have a defined wandering or deviating pattern as potential instances of disorientation. We develop a method called iBDD that is able to detect two categories of outlying trajectories in a uniform framework in real-time. Using ten individuals' real-world GPS datasets, we demonstrate that iBDD can achieve 95% detection rate of disorientation with less than 3% of false positives, based on properly chosen parameters.