Human action recognition using a semantic-probabilistic network

In this paper we propose a semantic-probabilistic network to recognise human actions. We use a predefined domain ontology to describe the events and scenarios in the scene as a hierarchical decomposition of simple concepts and variables and then perform an automated conversion of the ontology into a Bayesian network. A novel approach for Bayesian network nodes' weights calculation is introduced based on the weighted relation between concepts of the ontology in order to reduce the influence of incorrect object detection. We then evaluate the performance of our approach using it to predict gestures in a human gesture recognition system, using a set of pre-recorded video sequences.