+Istituto di Informatica
Università di Ancona - via Brecce Bianche - 60131 Ancona - ITALY
{tascini@inform.unian.it, puliti@inform.unian.it, zinga@inform.unian.it}
In this paper we consider the problem of the recognition and tracking of moving objects from monocular sequences of images. In particular, we describe a new strategy for 2D intelligent temporal tracking of objects. The approach is feature-based and resolves the correspondence between points in two consecutive images with geometric tracking. While in literature often they are used edges or corners here we use an advanced segmented image, that may be represented with a series of polygons, whose correspondence in consecutive images constitutes the basis of the tracking. The architecture of our vision system is structured on two sub-systems: the 'Static Recognizer', which allows to verify the presence of an object in a scene, and the 'Dynamic Recognizer', which allows to track an object after the relative translation and/or rotation.