Automatic Surveillance Using Omnidirectional and Active Cameras

We are developing  a real-time automated surveillance  system  which uses an omnidirectional video camera  in combination with  multiple  active  cameras.  Tracking  of  multiple subjects  in an  indoor environment  is  performed  using  omnidirectional  video  as  input.   The  world  coordinates  of  each subject  in the  room  are  estimated in  order to  direct the  attention  of  one  or more  pan-tilt-zoom  cameras.   The   system   automatically   controls   these   cameras   for   the   purpose   of   obtaining   high resolution  images  and video sequences  of subjects.  In particular,  we demonstrate  that high-quality facial images may be extracted from the images captured by the system.  The automatic  acquisition of  such images makes this system useful for experiments involving face and action recognition.