Our overall goal is to extract simple and reliable features which are descriptive, i.e. can be used by an unsupervised algorithm to discover the important image features in a large video set in order to detect unusual events.
