This prototype deliverable presents the design and recording of the Siren database. The rationale behind recording an audiovisual database of people playing games is that facial expressivity in the particular framework is different to that in general human-computer environments (where most of the existing audiovisual databases have been recorded in). While facial feature extraction and tracking algorithms can be adapted to work for gamers, the machine learning part (recognition of facial expressions) has to be trained from scratch, using the particular emotion classes relevant for gamers. As a result, recording people while playing and annotating the database was of utmost importance for the natural interaction component to provide estimates of player satisfaction and attention.
|D4.1 - Natural interaction and game play – first version (M10)||349.93 KB|