Semantic Object Recognition Based on Qualitative Probabilistic Spatial Relations

Abstract

Intelligent systems able to perform everyday tasks in human living environments are going to play an important role in the future. Especially service and assistive robots, which could take over tasks such as fetch and carry, would be of great use. However, dealing with objects in natural environments is not a trivial but rather very challenging task. The robot has to extract the objects from noisy sensor data and give them a meaningful and correct description. The goal of this thesis is to develop an approach for robust semantic object recognition, which can be used for such purposes. In our approach, we take advantage of the spatial contextual information about objects’ co-occurrences to perform a robust object recognition. Our approach is unique in that it uses spatial semantics in a probabilistic manner. We also develop a new representation of this information, termed Spatial Potential Fields.

This work was supported by the Graduate School SyDe, funded by the German Excellence Initiative within the University of Bremen’s institutional strategy.