A real-time scene change detector for the review of video images

A small, simple image processing system is presented which enables the detection of scene changes in real-time sequences of video images. Such sequences may be produced by a video camera during on-site surveillance or be played back from a video cassette recorder for reviewing. The system detects scene changes by reading pixel profiles along some user defined polylines and cross-correlating them with the corresponding profiles of a reference image. If the correlation factors of a specified set of profiles drop below their thresholds, an alarm is triggered and the reference image is updated. The cross-correlation technique makes the detector rather insensitive to changes in scene illumination. A prototype system, running on an IBM-PC/AT with plug-in image processing board, has been successfully tested on video images recorded in a plant