We are actively deveoping unsupervised techniques to discover human activity in video. Our approach relies on three coupled innovations. First, manifold representations of video actions that facilitate useful comparisons between activities. Second, a rapid site specific setup model where activites are learned quickly off a few examples and with minimal human direction. Third, underlying the first two are new ways of grouping and retrieving examples of video activity. A demonstration video is available on Bruce Draper's Homepage.
We are active in face recognition with a recent emphasis on recognition in video. Much of our work has centered on evaluting face recognition algorithms. We've been part of a series of efforts associated with NIST including FRGC, FRVT, ICE, MBGC and GBU. Our most recent efforts are centered around the Point-and-Shoot Face Recognition Challenge. This includes the FG 2015 Video Person Recognition Evaluation.
A key bit of recent work on face recognition is the The Point-and-Shoot Face Recognition Challenge (PaSC)