Proactive CCTV surveillance typically involves monitoring video streams from many different cameras trying to identify and react to incidents as soon as they appear. Computer vision may assist in the real-time processing of the video streams to draw the operator attention to events that may be of interest and thereby reduce the work load and increase the response time. Vision may also be used off-line to process previously captured footage to localise and track suspect vehicles or persons over a period of time. Applications range from identifying hooligans at a football stadium to detecting terrorists at long range as they approach a military facility or nuclear power plant.
Example application: Automatic forensic video retrieval
|In forensic video retrieval, CCTV footage is processed to extract features and build up a video index. When an incident has occured, investigators can search through the footage by asking queries based on the available features. Example features that may be automatically extracted by the video retrieval system are object shape, size, colour and type. It may also contain temporal trajectories of objects over time, object identity if available, and object activity. An example query could be "Return all footage with a red car on 49th street, at the corner with the 7/11 store". A video retrieval system can save many man-hours as browsing through footage in a large network of surveillance cameras is extremely time consuming.|