Abstract
An infant's unhealthy lifestyle increases the risk of adult diseases and the cost of medical care in their future. If the children's life-log is recorded and then their physical condition and transition of activities can be analyzed, it is expected that information can be utilized for the prevention of such issues and for promoting good health. Therefore, activity recognition for infants is discussed in this paper. Feature points are computed from accelerations measured using a 3-axis acceleration sensor attached to the infant's upper-arm, and then, a two-phased classification is performed. In addition, we developed an activity recognition system with database functions and real-time relearning for infants.
