Robot mapping, both indoor and outdoor, is typically based on sets of 3D measurements of the environment (point clouds) coming from either laser range finders or RGBD cameras. While both of these sensors provide accurate data about objects within a relatively wide range, they fail to provide directly informative readings about transparent or highly reflective objects, which are commonly found in cluttered indoor environments such as homes and offices. This paper describes a method of recognising that there are transparent objects within a scene and reconstructing them from the limited information that is available. Our method is based on reconstructing geometric properties of the missing objects using inference from the shadows that are left. This provides an estimation of the volume of missing objects. We demonstrate the methods first on regular measurable object to compare our estimation with measured data and present the reconstruction of two exemplary transparent object