Tracking deformable objects in geospatial applications

In this paper we describe a novel approach for change detection of moving deformable objects. We assume that each object is represented as a closed polygon (convex or concave). First, we use differential snakes to track the object outline between two different frames. Next, we examine the geometric properties (area moments and principal axes) of these two polygons and we estimate their changes (translation, rotation, area expanded or contracted). Additionally, using results from invariant moments theory we compute a similarity index among different instances of an object's history. This information is used to update geospatial databases and to support complex analysis. Experimental results on real imagery illustrate the capabilities of our approach.