A state preserving approach to recognizing human behavior using wireless infrared and vibration sensors

Some home monitoring research has tried to deal with a broad range of human behaviors such Activity of Daily Living (ADL) or Instrumented ADL (IADL) in an entire home, but most research results does not yet seem satisfactory. So we focused on modeling human behaviors in a bedroom, not an entire home. To do this, we define a behavior state and behaviors. We constructed the experimental system by installing a wireless infrared sensor on the ceiling above vibration sensors located on a bed. Based on the behavior definition, we formed a feature vector for each behavior in a bedroom. The total 110 experiments on five behaviors were tested by 5 subjects. We achieved a recognition ratio of 75.4%. After analyzing causes of false negatives, we revised some features. Then, the total 115 experiments were tested by four additional subjects. The recognition ratio improved to 94.8%. The high recognition ratio indicates that the proposed behavior model is effective in recognizing human behaviors in a bedroom. This research will contribute to forming a foundation to track human behavioral log in a bedroom.