Improving Object Detection and Recognition for Semantic Mapping with an Extended Intensity and Shape based Descriptor

We propose BASE, an extended descriptor for RGB-D images, that efficiently combines intensity and geometrical shape information to improve discriminative power. We use this new descriptor to detect and recognize objects under different illumination conditions and apply it within an adaptive boost classification framework to provide semantic information in a mapping task. We compare the performance of our descriptor against two standard ones in the literature. Experimental results show that in spite of the simplicity of the descriptor and of the Adaboost training approach, high accuracy classification is obtained with fast processing time.