Defining and evaluating qualitative spatialrelations for describing desktop scenes

Abstract: In an everyday scene such as an office, one can find lots of objects with different sizes and shapes. Normally, these objects are placed following a certain order. Describing the different spatial relations between the objects in an environment is a simple task for a human being but it is not for a robot. This is due to the nature of these relations being abstract and this requires certain degree of intelligence in a robot. This thesis investigates ways for a robot to find descriptions in a scene using spatial relations such as “the mouse is to the right of the keyboard”. To deduce these relationships, accurate information about the position and orientation of the objects is needed and to achieve this purpose a software was developed. In order to use this information to extract descriptions of a scene, a system was implemented to quantify to what degree certain relations hold. Finally, several scenes were analyzed, by obtaining its respective description, trying to extract some patterns.