DETERMINATION OF APPARENT MOBILE AREAS IN AN IMAGE SEQUENCE FOR UNDERWATER ROBOT NAVIGATION

Abstract

This paper deals with the determination of moving areas in an image sequence from the variations in time of the intensity distribution. Beyond this problem we are concerned with an application whose aim is the estimation of the trajectory of an underwater vehicle by means of a video sensor on the robot pointing to the sea bottom. The trajectory will be piecewisely reconstructed by tracking a series of appropriate elements, i.e. apparent moving areas. throughout the image sequence. Contextual information is explicitely introduced in the moving-area detection-and-recovery process, which is stated as a statistical labeling one. The label field is modeled as a Markovian field using Gibbs distributions defined on a spatio-temporal neighborhood system. A solution to this labeling problem is formulated according to the maximum a posteriori (MAP) criterion. We have adopted a deterministic iterative algorithm to optimize the criterion at hand. Experiments with an underwater video image sequence are presented.