SEARCH IN LARGE VIDEO ARCHIVES USING QUERIES BASED ON A SYNTHETIC 3D HUMAN MODEL
Stephan Veigl, Csaba BeleznaiAbstract
We present a demo of a query concept to facilitate and improve the search for a specific person in large video surveillance archives. The query is defined as a combination of a conventional rule-based system and a synthetic human model or avatar. The query is enabled by a framework which includes a graphical synthesis tool capable of generating images based on a 3D human model capturing pose, view and a range of appearance variations. The generated synthetic image is used in a query-by-sample fashion to perform the search. A highly optimized implementation utilizing GPU and parallel CPUs allows a fast response on large archive datasets (hours of videos) and permits effective appearance-specific search and browsing.
Read Submission [892]