ViHASi: Virtual human action silhouette data for the performance evaluation of silhouette-based action recognition methods

In this paper we introduce a large body of virtual human action silhouette (ViHASi) data that we have recently generated for the purpose of evaluating a family of action recognition methods. These are the silhouette-based human action recognition methods. This synthetic multi-camera video data-set consists of 20 action classes, 9 actors and up to 40 synchronized perspective cameras. This data-set has been recently made available online for other researchers to download. In order to demonstrate the usefulness of the ViHASi data we make use of an existing action recognition method that is simple and relatively fast. Moreover, to deal with long video sequences containing several action samples, a practical temporal segmentation algorithm is introduced and tested that is tightly coupled with the action recognition method used. The experimental methodologies outlined here provides a route towards quantitatively comparing silhouette-based action recognition methods.