Semantic Place Labeling Method

The paper presents a method of semantic localization of a mobile robot. The robot is equipped with a Sick laser finder and a Kinect sensor. The simplest source of informa tion about an environment is a scan obtained by the range sensor. The polygonal approximation of an observed area is performed. The shape of the polygon allows us to distinguish corridors from other places using a simple rule based system. During the next step rooms are classified based on objects which have been recognized. Each object votes for a set of classes of rooms. In a real environment we deal with uncertainty. Usually probabilistic theory is used to solve the problem but it is not capable of capturing subjective uncertainty. In our approach instead of the classic Bayesian method we proposed to perform classification using Dempster-Shafer theory (DST), which can be regarded as a generalization of the Bayesian theory and is able to deal with subjective uncertainty. The experiments performed in real office environment proved the efficiency of our approach.