An activity/localtion/time reminder system is able to take into account the user's activity state independent of time and location. This not only allows reminders to be triggered by activity (e.g. "If I'm speaking too loudly, remind me to calm down") it also allows the system to balance overall demands on the user's attention. For instance, such a system might follow the rule that reminders are not to be delivered while the user is is engaged in a high-attention task like talking on the phone, or a high-risk task like crossing the street.
The difficulty is that recognizing user actions (or environmental conditions) generally requires high-bandwidth sensing and sophisticated analysis. Computer vision techniques have been shown to be effective in identifying user location and action [5,1], but only recently has it become possible to do this type of sensing and analysis in an easily portable or wearable package, independent of outside infrastructure, with reasonable battery life and performance.
The goal of the Memory Glasses project is to produce an activity/location/time based proactive reminder system that is powerful enough to recognize a wide variety of user activities and environmental conditions, is flexible enough to be trained to recognize new user activities and conditions easily and with a minimum of user intervention, is ergonomic enough to be worn while engaged in a wide variety of daily activities, is reliable enough to be trusted, and makes the best possible use of the user's time and attention. In the next section of this document we outline the architecture for the Memory Glasses prototype system as a work in progress.