Video mosaicking along arbitrary vehicle paths

Abstract

This paper presents the experimental verification of a novel method for real-time video mosaicking of the ocean floor from a semi-autonomous underwater vehicle. In order to enable mosaicking along unconstrained vehicle paths, it is essential to reduce the propagation of image alignment errors within the mosaic. We have developed iterative smoother-follower techniques to reduce these errors, and we have proven their effectiveness in the laboratory environment. We also present our results from a concurrent research effort to demonstrate a completely autonomous video mosaicking mission. By transferring our vision technologies to OTTER (Oceanographic Technologies Testbed for Engineering Research), we have created a prototype system for visual survey of the sea floor. In particular, we discuss the development of a new vision processing subsystem and its integration into the AUV control system.