Building 3D elevation maps of sea-floor scenes from underwater stereo images

Abstract

Stereo vision has been used in the application of terrestrial mobile robots and the Mars Rover to construct a three-dimensional map of the environment from two-dimensional images. The geometrical issues of the problem for both calibrated and uncalibrated systems have been well studied. However, the main problem difficulty is solving the correspondence problem, or equivalently the accurate computation of the disparity, between the left and right images. The application of binocular vision to underwater imagery and some of the problem complexities arising from the inherent characteristics of the environments are investigated. Results of an experiment with underwater data under three turbidity conditions are presented to demonstrate some of these issues.