Enhancing multimedia search using human motion

Over the last few years, there has been an increase in the number of multimedia-enabled devices (e.g. cameras, smartphones, etc.) and that has led to a vast quantity of multimedia content being shared on the Internet. For example, in 2010 thirteen million hours of video uploaded to YouTube (http://www.youtube.com). To usefully navigate this vast amount of information, users currently rely on search engines, social networks and dedicated multimedia websites (such as YouTube) to find relevant content. Efficient search of large collections of multimedia requires metadata that is human-meaningful, but currently multimedia sites generally utilize metadata derived from user-entered tags and descriptions. These are often vague, ambiguous or left blank, which makes search for video content unreliable or misleading. Furthermore, a large majority of videos contain people, and consequently, human movement, which is often not described in the user entered metadata.