Hierarchical tracking of motion in multiple images

A hierarchy of feature types are employed to detect motion, establish temporal tracks and fuse motion information from multiple cameras. The development of such a hierarchy has been motivated by applications which require the tracking of objects in open space environments such as car parks or shopping malls. Such applications include security monitoring of sites, the monitoring of vehicles entering, manoeuvring, parking and leaving carparks, and the generation of marketing data in shopping malls from the automatic analysis of shoppers' browsing behaviour. Establishing correspondences between features in images is difficult as low level features tend to exhibit stable statistics over very short intervals and may temporarily disappear because of occlusion and low contrast with the background. The tracking process may be improved and a high degree of long term motion stability obtained by introducing the concept of hierarchy. In addition, fusing information from different viewpoints is more appropriate at higher data representations. The paper describes a hierarchy of features used to establish a scene description.