This thesis presents a method to compute a volumetric model of an arbitrarily-shaped convex object, immersed into a transparent medium inside a transparent container, from multiple views. Different views are generated by rotating the object in front of a stationary camera using a turntable. Intrinsic camera calibration is done using different views of a chessboard pattern. Extrinsic camera calibration is done semi-automatically using images of a chessboard pattern rotated by the turntable. The images of the transparent container are taken separately at identical rotation angles. After the side walls of the container have been automatically detected using these images, they are used to model the refraction. Refraction is included into the reconstruction algorithm Shape from Silhouettes by incorporating the refraction into the projection model.