Abstract
The construction of 3D models from images is a common task in Computer Vision. Mutli-view stereo and structure-from-motion methods have advanced to the point where 3D models of many types of objects can be easily made from images captured with consumer-grade cameras. Some objects, however, are hard to image in their entirety. In this paper we consider flat objects - those with one dimension much smaller than their other two. Such objects are easy to model from either side, but creating a full 3D model is challenging. Our approach projects the two half-models into two dimensions, finds their convex hulls, and then aligns them to merge the two shapes. We demonstrate this method on an archaeological application - modelling the waste flakes produced during stone tool manufacture. We show that our method can create complete 3D models from images captured of the ventral and dorsal (front and back) sides of the flakes. Comparing these models to reference laser scans we find that whole flakes can be reconstructed with between 0.5 and 1.5mm average reconstruction error.