When service robots carry on mission planning and execution, they not only need structural information about the environment to navigation and localization, but also need to possess deeper knowledge to endow a robot with higher degrees of autonomy and intelligence. However it is hard for a robot to cognize the semantic description such as the function attributes of objects and rooms only relying on robot’s sensors itself. When the exploration scope of the robot is expanded to large regions such as a building, the problem becomes more prominent. In this paper, artificial labels are used to provide semantic concepts and relations of objects and rooms in a building, which can solve the complexity and limitations of semantic recognition and scene understanding only with robot’s vision. Imitating spatial cognizing mechanism of human, three hierarchy space model is built for a building service robot. With this space model, not only the plane structure of the environment for navigation is known, but also the abundant semantic information in different kinds of environment are obtained. Simulation results show that the three hierarchy indoor space model is applicable to service tasks of indoor robot, understanding human semantic statement, produce reasonable service path, and ensuring the security of robot running in building environment.