Research is proposed on innovative solutions for building a network of pan-tilt-zoom surveillance cameras without overlapping fields of view that can intelligently calibrate and localize themselves and in relation to each other. Currently, such systems typically consist of a central recording center, with hundreds of hours of tapes recorded daily. There is usually no coordination between cameras, other than what is provided by human operators at the center. Existing research in surveillance is primarily focused on classical and highly studied problems such as tracking, monitoring, and recognition. Little attention has been paid on the actual coordination of cameras in a network. This lack of attention to an important aspect of surveillance systems presents opportunities for discovery that will be explored in this project. The proposed solutions to these problems will significantly advance the state of the art. Self calibration of cameras will use the vanishing line of the ground plane and the vertical vanishing point obtained by tracking people over multiple frames. Ego-motion of cameras subject to pan, tilt or zoom will be accurately measured by considering invariant trajectories in space- time volumes of video data. Relative positions between cameras, which need to be computed only once, will make use of commonly seen points of the scene if cameras can be panned to have temporary overlaps of fields of view, and survey points in the scene if overlaps are not possible. The main outcomes of this project will be (1) a set of innovative algorithms that could transform the current approach to wide area surveillance, and move it one step closer towards automated operation, and (2) an actual implementation and operational evaluation of these algorithms with real surveillance systems.
Surveillance issues that could benefit from this effort are target tracking, detection of actions, identification of activities, and recognition of tracked objects and people of interest. The implications and the impact are therefore across all application areas in surveillance, but more specifically in wide area surveillance, e.g. monitoring activities in an airport or on a campus. Wide area surveillance is a challenging research topic, since it requires orchestrated coordination of cameras.
The proposed research is an important step ahead in an area related to homeland security. The graduate students involved in the project will benefit from research activities in an area of growing importance and national priority; in addition, the PI is preparing a graduate-level course on "vision-based surveillance".
URL: http://cil.eecs.ucf.edu/
