Invisible and Cost-Effective Sensors with a Network Robot for an IoT House for Tourists

This paper presents a novel application of our previously developed sensor system to recognition of behavior patterns using a network robot and cost-effective invisible sensors. We set up the system at an actual house as a concept of an internet of things (IoT) house for tourists. We obtained 14 person-day benchmark datasets from ten people in their 20s. For constructing benchmark datasets, they recorded event times of seven patterns: getting up, sleeping, going out, coming home, emergency calls, opening or closing of a refrigerator, and the use of a TV remote control. As ground truth labels for cross-validation-based evaluation, we integrated them into three patterns: going out, staying at home, and sleeping on the bed. The experimentally obtained results revealed that the mean recognition accuracies with random forests were 99.60%, 99.30%, and 98.54% for the respective three datasets.