A dynamic vision system for tracking and localization of underwater objects

Abstract

We describe the development of an underwater cross-range high-accuracy 3D metrology by utilizing a pair of video cameras for underwater vehicle applications. The novelty of the approach lies in (1) a polynomial representation of image non-linear distortion induced by lens system and surrounding water; (2) the transfer of photogrammetric bundle adjustment techniques to underwater ranging; and (3) a new representation of underwater stereo vision including camera self-calibration, match points searching, and computation of the target coordinates. We demonstrate that our method achieves better flexibility and accuracy compared to traditional stereographic methods.