Model-referenced underwater localization using optical and acoustic images for autonomous intervention systems

Abstract

This thesis deals with a localization method for autonomous intervention systems that autonomously perform underwater intervention tasks. This method uses image information from optical and acoustic camera to provide high-precision localization information necessary for the intervention systems to maneuver near the intervention environment and perform the precision manipulation on target structures.

The thesis proposes three localization techniques according to the configuration of the mounted sensors of the intervention systems. First of all, a localization method using optical images for the intervention system is introduced. This method estimates relative pose of six degree of freedom through visual referencing of the object of interest in a single optical image. The method considered edge features for robust visual referencing in underwater images, which use three-dimensional wire-frame model for the intervention environment. In addition, a localization method using acoustic images for the intervention system is proposed. For a single acoustic image, the thesis sets the localization problem as an optimization problem to find a six-degree of freedom pose hypothesis that produces the most similar acoustic image in the acoustic camera simulator. To determine initial inputs for the optimization problem, the thesis proposes an image nearest neighbor search that determines a pose hypothesis as an initial localization from image-pose hypothesis group generated in advance using the acoustic camera simulator for a space of interest. In this regard, this thesis describes a method of constructing the acoustic camera simulator that efficiently generates simulated acoustic images. This thesis proposes a strategy for effectively estimating relative poses using optical and acoustic images in consideration of the operating and measurement characteristics of the optical and acoustic camera. This strategy combines two localization methods, which estimates the relative pose using acoustic images as the main information and selectively uses the optical images according to the quality of the underwater image.

This thesis describes how to build an efficient localization system using a filter algorithm by fusing underwater navigation sensor information and the localization information from each localization technique. Finally, the thesis looks at the performance and usefulness of the proposed techniques through simulation and experimental verification and discuss the results.