One theme is the recognition of an individual at a distance through gait and face information. 
Current research on recognition through gait ranges from cooperative approaches (with markers, e.g. at ankles), using simplifying assumptions (homogeneous background, fronto-parallel motion) to the recognition of specific gait types (running, walking, jumping, etc.). A general, non-cooperative approach to the identification of persons in unconstrained environments has not been developed. We are investigating various methods of non-cooperative gait recognition using both 2-D and 3-D human models. Probabilistic approaches will be used to integrate temporal changes. View invariant gait recognition will be investigated. Furthermore, most present face recognition approaches assume the facial images to be at least 100 pixels in width. The use of small-scale facial images of width between 70 and 120 pixels has not been attemted. A probabilistic framework will be used to cope with small scales as well as severe scale changes over time. 

The second major theme of our research is the evaluation of different scenarios that support human identification at a distance. Those scenarios will use near-IR cameras and a set of actively controllable 5-degree-of-freedom cameras (the degrees of freedom are pan, tilt, zoom, shutter speed and iris). Evaluations will allow us to determine the best recognition results that can be achieved, how various data types (color cameras, IR-cameras) can be fused, what viewing angles are best, whether cameras should be synchronized, and to what extent camera calibration is possible and needed. 
 
 

Home | Overview | People | Research | Publications | Data | Links | Contacts                                                     Site Last Modified: 4/17/2001