Real-Time Labeling of Places using Support Vector Machines

Humans refer almost to everything by their characterization rather than their detailed descriptions. For example, in indoor environments places are specified as: rooms, corridors, etc. Such categorizations, if learned by a robot, could improve the capabilities in the areas of navigation, localization, or human- robot cooperation. This paper studies the problem of categorizing environments into semantic categories. A new approach based on Support Vector Machine (SVM) is proposed and described for learning to perform classification of environment. The SVM is trained using a supervised training algorithm. This method uses simple features extracted from laser range measures, using methodologies normally used in computer vision. In the present paper the proposed method is used to distinguish between two classes of places from sensor data: rooms and corridors. The real-time experimental architecture designed for classification is presented. Experimental results obtained with real sensor data demonstrate the feasibility and effectiveness of the proposed approach.