Abstract: An important sub-problem of visual action recognition is identifying the roles of individuals in interactions with multiple individuals when viewed from a camera. This project aims to infer person roles based on semantic interactions, a critical requirement in real settings where individuals' identities must not be identified. The proposed system should enable the identification of people roles by associating them with appearance features, inferred interactions and detected activities (i.e. semantic motion dynamics associated with each of the observed roles). Dataset: develop datasets (e.g. game simulated data).

Contact: David Ahmedt-Aristizabal