Surveillance and tracking in feature point region with predictive filter of variable state

Surveillance in today's world is a very common issue in computational vision. This activity is present in literature in two different ways: first, as having both camera and objects in motion (Behrad et al. 2000); second, having detection of moving objects by means of one static camera (Lipton et al. 1998). This paper is centered in the last approach, where the interest is to find the movement of objects in images by detecting temporal differences and to define the movement region, which is analyzed by growing region, selecting one region and tracking the object. Once the region is selected, the interest points are determined through a modified corner detector of Harris et al. (1988). A reference data bank is created, to be used in the matching process and determining the characteristic of corresponding points. With these corresponding points, the movement parameters of the region can be estimated and the prediction filter (VSDF) in the tracking cycle initialized. The method that is developed here consists in considering the tracking cycle a matching process by normalized correlation with the help of the prediction filter to adjust the estimated measurements. Thus a method that allows tracking of points of interest in a surveillance region, in a stream of images with significative results to implement appropriate real time algorithms. In this stage of our research Matlab and regular digital cameras were used for prototype design of tools and experimenting.