Tracking deformable objects in geospatial applications

Abstract

I n this paper we describe a novel upprouch f b r change detection of moving defrmable ohjects. We o.wiime that each object is represented us a closedpolygon (conver or concave). First, we use di,@entiul snakes Io track the object oirtline between two different frames. Next. we examine the geometric properties (area moments, and principal axes) of these hvo polvgons and we estimate their c~hunges (translation, rotation, area erpanded or contracted). Additionally, using rcm1t.s from invariunt moments theory we compute a similarity index among di@erent instances of an ohject ' . s histo?. This infirmution is used to npdute geospatial datuhases und to siippart comp1e.r analysis. Experimental results on real imagery illiistrulr the capuhilities of o w approach.