Abstract
New generation of 2-D sonar cameras operating at 1-2 MHz provide images with enhanced target details in highly turbid waters, although range is reduced by one to two orders of magnitude compared to traditional low-/mid-frequency systems. Therefore, they are suitable imaging systems for the short-range inspection of underwater structures. As for 2-D optical images, multiple object images from nearby viewing positions may be utilized for 3-D shape reconstruction based on similar visual cues as the motion parallax. In this paper, we address the reconstruction of 3-D points from two sonar views, acquired either simultaneously by two cameras, or from a single camera at two known relative positions. We investigate a number of linear algorithms for 3-D reconstruction from matches in two views, and examine some degenerate configurations. While these do not provide an optimal solution, e.g., in the Maximum Likelihood sense, they can offer a good initial condition to ensure effective convergence of the ML estimate. We present results of experiments with synthetic and real data in support of our theoretical contributions.