A coherent semantic mapping system based on parametric environment abstraction and 3D object localization

An important prerequisite for intelligent robots to effectively perform daily tasks in indoor environments is an advanced environment model. Not only should this model contain metric level information about the geometry of the perceived environment, it should also provide abstract level information, such as topology and objects, so as to benefit high level robotic applications. In this paper, we propose a mapping system which generates a coherent semantic map for indoor environments while taking laser range data and RGB-D images as sensor input. We propose to realize this system by combining parametric environment abstraction, i.e. using a parametric model to approximate the geometry of the perceived environment, with 3D object localization. Experiments using real world data show promising results and thus confirm the usefulness of our system.