Stereoscopy and Scene Reconstruction

Abstract

Depth perception is of great use to humans in many situations, therefore it could be very useful for artificial systems to be able to use this capability in their many applications. One of the most common means of deriving depth information from two images is stereoscopy, or stereopsis. In this bachelor thesis project a novel approach to computational stereoscopy is introduced in the form of a two-stage model. First a very fast, but inaccurate form of template matching (“Sloppy Matching”) is performed, which is then improved upon by an iterative quasi-physical simulation to remove erroneous matches (“Corrective Smoothing”).