Recognition of three-dimensional objects from orthographic views

This paper presents a program interpreting orthographic drawings input through a digitizing tablet into three-dimensional (3-D) objects. Ambiguity arises in such a task not only in the case of two views but also with three-view inputs, due to coinciding projections of vertices. The main effort consists in heuristically reducing this ambiguity. The general strategy is a bottom-up identification of the objects starting with the tablet-coordinates and aggregating them into vertices, then faces and finally polyhedra. The program does not contain any high level geometrical concepts or probable properties of 3-D objects but only some basic definitions of polyhedra (as expressed by Euler and Mobius). By applying these definitions to different parts of the scene to recognize, it defines local sets of alternatives and then chooses the right ones by running a mini theorem prover over these sets. A major issue in writing this program has been to limit the drawing constraints imposed on the user. In case that it gets blocked in the recognition process, however, the program can require the user's help.