Multi-camera person tracking system

For the surveillance of the large interior spaces it is often necessary to use multiple cameras. This paper deals with the tracking of an individual using multiple cameras with partially overlapping views. To implement such a system it is necessary to preprocess recordings, detect a person in motion and find the contact point between person and floor. It is especially important to consider the situation in which a person is in the folded multi-camera field of view. Last step is making homographic transformation of a point where person contacts the ground in observed space. System performance was measured on artificially generated recordings of a person moving through halls. It utilizes views with cameras from four different locations. Implemented application runs fast enough for use in systems that operate in real time.