Video surveillance tracking using colour region adjacency graphs

An automatic tracking system is presented which uses the matching of attributed region adjacency graphs to track moving objects. The problem of object recognition with complex colour images is always made difficult by the changing shape and illumination of objects moving relative to complex light sources. Moreover, a moving camera causes additional complications in distinguishing an object. However, by identifying the colour regions of an object in one frame and looking for these regions in the next frame, object tracking can be achieved. This colour segmentation is efficiently performed within a novel 2 1/2 dimensional colour space drawing on established Hough transform techniques. Automatic visual control of the motion tracking camera demands real-time processing of video streams.