Optimal search mapping among sensors in heterogeneous smart homes

Abstract

There are huge differences in the layouts and numbers of sensors in different smart home environments. Daily activities performed by residents trigger a variety of sensor event streams. Solving the problem of sensor mapping is an important prerequisite for the transfer of activity features in smart homes. However, it is common practice among most of the existing approaches that only sensor profile information or the ontological relationship between sensor location and furniture attachment are used for sensor mapping. The rough mapping seriously restricts the performance of daily activity recognition. This paper presents a mapping approach based on the optimal search for sensors. To begin with, a source smart home that is similar to the target one is selected. Thereafter, sensors in both source and target smart homes are grouped by sensor profile information. In addition, sensor mapping space is built. Furthermore, a small amount of data collected from the target smart home is used to evaluate each instance in sensor mapping space. In conclusion, Deep Adversarial Transfer Network is employed to perform daily activity recognition among heterogeneous smart homes. Testing is conducted using the public CASAC data set. The results have revealed that the proposed approach achieves a 7–10% improvement in accuracy, 5–11% improvement in precision, and 6–11% improvement in F1 score, compared with the existing methods.