Automated video-surveillance systems must employ effective
low computational cost algorithms in order to process alarms in
the greatest number of cameras possible with satisfactory and
real-time results. There are a lot of works based on multi-camera
systems which offer solutions with solid results but do not take
into account the processing time and only can manage a reduced
number of cameras in real-time, as is studied in Wang (2013).
Our approach, focused on a complete surveillance of a shopping
mall, can operate in a naturalistic multi-camera model efficiently,
even in difficult conditions originated by video compression and
low quality images derived from a previous data transmission to
the control center. This efficiency allows managing several cameras