Please choose your delivery country and your customer group
One of the key challenges in surveillance video face screening against a Watch List is the fact that faces in surveillance video are often observed at an angle different from the angle at which the faces are captured in the Watch List. Particularly, facial images in surveillance video are normally observed at various pose angles, and from above eye level. In contrast, mugshot (reference facial images) stored in databases are regularly captured at a frontal post and at eye level, thus causing poor matching between the images. One way to overcome this problem is seen in advanced preprocessing of stored images. It is possible to synthetically generate variations of a reference facial images of target individuals at under the same conditions (e.g. pose angle) under which they will be most likely observed in a video. While several commercial tools exist, an open source library is available to generate a 3D face model from arbitrary 2D facial images. This library, called Candide, may allow academia and industry to significantly improve the matching performance of their algorithms in video surveillance applications. This report overviews this library and analyzes its suitability for the problem.