Localization of a fleet of AUVs using visual maps

The present paper describes a vision system for the construction of a visual map and motion estimation for a set of AUVs (Autonomous Undersea Vehicle). Robots are equipped with down-looking camera and sensors which are used to estimate their motion with respect to the seafloor and to built real-time seafloor maps. Each robot computes relevant images and sensor information to send to undersea central station responsible for the map construction. This paper use two approaches to built maps: a mosaic process using an Augmented State Kalman Filter; and a Self-Organized Maps using Kohonen networks. The theoretical concepts associated with both are presented, and results are discussed.