Volume intersection with optimal set of directions

Abstract The Volume Intersection algorithm for 3D reconstruction has been making use of a set of prespecified directions. In this work, we address the problem of determining the optimal viewing directions during the process of reconstruction. As the problem is intractable, we propose a heuristic which is demonstrated to work with better accuracy than earlier methods of prespecified directions. For a given silhouette, its width determines two orthogonal directions, one along the lines of support with minimum width and the other perpendicular to these. These two directions reveal maximum surface area of the object and also restrict the width of the silhouettes taken along these new directions. Based on this principle an iterative algorithm is proposed in this paper. Experimental results with this heuristic with simulated objects are reported. The objects are reconstructed from three views, i.e., front view and the two optimal directions obtained from the front view. This reconstruction is compared with that from front, top and side views. In most cases, our method gave better results in terms of accuracy.