Skip to Main Content
Learning human daily behavior habit patterns from sensor data is very important for high-level activity inference of service robot. This paper proposes a model that represents person's daily behavior habit pattern. Firstly, a coordinate frame is defined on a map built by mobile service robot, and two key variables are calculated using consecutive data collected by the robot. Then, based on two key variables and the states that are defined in advance, the probability model is built. In order to learn the model efficiently, EM algorithm is applied. Experiment results demonstrate that the model is feasible to learn human behavior habit and can afford a judging gist to detect persons' unwonted behavior.