Precise mosaicking of underwater images is essential for Mine Counter-Measure (MCM) operations. Inaccuracies in the navigation of the Autonomous Underwater Vehicles (AUV) that acquire the images may complicate the construction of the mosaics and cause ambiguities in the total number of targets and their positions. In this paper a method for automatically mosaicking a set of partially overlapping image tiles is presented. The method uses a robust implementation of image correlation that can cope with different illumination directions. Constraints to the relative positions and orientations of pairs of overlapping tiles are derived from the local correlations, and later used to drive a global optimization method that models the constraints as mechanical forces. Modeling the tiles as rigid rectangular plates, the dynamic evolution of the mechanical system will converge to an equilibrium configuration that corresponds to the optimal solution to the mosaic construction problem.
展开▼
