The annotation scheme of the MOBOT dataset

Abstract

Previous work in the framework of the MOBOT European project has led to acquisition of a multimodal sensory corpus that is intended to become the primary source of data retrieval, data analysis and testing of the mobility assistive robot prototypes currently under development within the project. The rich and multi-layered annotation of this corpus is a key factor for the usability of the acquired dataset. To this end, in this paper we report on the definition of the adopted annotation scheme, focusing on annotation tiers applied to audio-gestural signals intended to provide the ground truth dataset for the training and evaluation of visual and audio recognition algorithms, and also form the basis for the definition of the project's multimodal human-robot communication model.