Co-operative computing for a distributed network of security surveillance cameras

This paper reports on current research to build a computer vision system using a network of intelligent cameras to detect and track targets (i.e. people) in man-made environments over a wide area monitored by multiple video cameras, such as within factory premises, office buildings, railway stations, prison compounds or shopping precincts. Targets are tracked throughout the duration of their passage through the environment, providing a complete history of their movement patterns for subsequent interpretation. Such information can then be made available to the security operators in the event that an illegal or abnormal activity is detected. This paper focuses on the requirement for delivering a stream of multiplexed video to a central video monitor station, based on the use of a simplified MPEG video encoding. This approach utilises the target detection process to identify sub-images in the image data. These rectangular image fragments are JPEG encoded and transmitted on demand to the central monitor station for display, where they are superimposed onto an image of the (static) background reference image