The orthographic projection method for the description of three-dimensional objects
Chung‐Lin Huang, Julius Τ. Tou · 1987
This dissertation presents an orthographic projection approach to automatic description of 3-D objects from 2-D intensity images. Given images taken from different viewpoints of the designated 3-D object, a system is developed to generate the same three orthographic projection views. This study mainly involves the generation of a unique model of the viewed object from its input image, followed by the verification between the stored model and the generated model. The intensity images through image processing techniques are converted to binary boundary images. The Hough Transform technique is improved to fetch the embedded features in the images which include the line segments, elliptical arcs, and ellipses with more accuracy and less computation. However, the rounded edges of the objects cause small intensity differences and cause some features to be missing from the images. Using the labeling technique, heuristic inference, and surface gradient properties, the system may recover the missing features and generate the pictorial drawings. The pictorial drawings are viewpoint-dependent. However, with the knowledge of the shape from contour, the gradient space, and heuristic inference, the system can develop the surface orientations of the object and recover its 3-D information in the viewer-centered coordinate system which is then transformed to the orthographic projection (O.P.) coordinate system. The footprints of a 3-D object consist of the top view, the front view, and the side view which are three orthographic projections of the objects on three projection planes in the O.P. coordinate system. There are two phases in the system: the model generation and object identification. In the first phase the user interactively enters the known objects and the system will automatically create their relational models and store in the database, whereas in the second phase the system not only generates the relational model of the viewed objects but also finds the best match between the generated model and the stored models to interpret the object.