Extensible Activity Recognition System for Behavior Support

The number of behavior-support services for user activities has been increasing lately. To provide a highly convenient service, it is necessary to recognize detailed activities of users. However, in reality, to provide a service in a real environment, some issues need to be resolved such as developing devices required for recognizing activity, handling the vast workload involved in creating features such as rules for detailed recognition of user activities, and dealing with an increase in the volume of log data to be transferred to systems for activity recognition. This paper proposes grouping technology of a micro-activity indicating user's simple behavior and locations that automatically constructs features to be used for activity recognition and a selective filtering technology that transfers only the necessary logs by using the connected graph of activity, thereby reducing the amount of data to be transferred. These technologies can reduce the workload that is involved in creating features necessary for providing a behavior-support service in a real environment and the amount of log data to be transferred. As a result of applying the technologies that were developed for the behavior-support service for an assumed white-collar business, we verified that a classification rate of 76.8% of user activity could be achieved, along with a reduction in the workload of over 70%, and a reduction in the log transfer amount of 30%.