More and more computer applications are using three-dimensional models for a variety of uses (e.g. CAD, graphics, recognition). A major bottleneck is the acquisition of these models. The easiest method for designing the models is to build them directly from images of the object being modelled. This paper describes the design of a system, MOLASYS (for MOdeL Acquisition SYStem), that allows the user to build object models interactively from underlying images. This would not only be easier for the user, but also more accurate as the models will be built directly satisfying the dimensions, shape, and other constraints present in the images. The object models are constructed by constraining model points and edges to match points in the image objects. The constraints are defined by the user and expressed using a Jacobian matrix of partial derivatives of the errors with respect to a set of camera and model parameters. MOLASYS then uses Newton's method to solve for corrections to the parameters that will reduce the errors specified in the constraints to zero. Thus the user describes how the system will change, and the program determines the best way to accomplish the desired changes. The above techniques, implemented in MOLASYS, have resulted in an intuitive and flexible tool for the interactive creation of three-dimensional models. / Science, Faculty of / Computer Science, Department of / Graduate