Refining 3-D object models constructed from multiple FS sonar images by space carving

A sequential space carving method has been developed for the 3-D reconstruction of objects from multitude of forward-look sonar images captured at known sonar poses [1]. The 3-D space within common viewing volume of several camera poses is divided into small volumetric pixels (voxels). Projecting each onto various images, all voxels satisfying certain consistency measure are maintained. Conversely, voxels violating the consistency measure are carved out as inadmissible. The 3-D space comprising of all admissible voxels is taken as the potential object interior. Finally, the boundary of this volume defines the object surfaces. The method applying a binary admissibility criteria does not make use of image intensity values from any one of multitude of available sonar views. Here, the refinement of generated 3-D surface model is explored based on optimization formulations that minimize the discrepancy between the sonar intensity measurements and predicted values by the sonar image formation model. Comparison among various formulations leads to a selected one with desirable properties for improving the surface model.