Recovery of structure from motion: Implications for a performance theory based on the structure-from-motion theorem

Abstract

The structure-from-motion theorem asserts that three projective views of four elements in motion are sufficient to specify the structure of the object to which the elements belong. Seven experiments were conducted to determine whether the perceptual recovery of structure in human observers is favorable to an interpretation of the theorem as a performance theory. In six experiments, subjects either discriminated simulations of objects in rotation around the y-axis versus randomly perturbed counterparts or tried to identify them. As the rotation angle between frames of the simulations increased, performance accuracy decreased. Furthermore, performance accuracy did not depend upon the type of projection (parallel versus polar) or upon the coplanarity/ noncoplanarity of the stimulus elements. A seventh experiment showed that the subjective appearance of structure breaks down when the angle of rotation between successive frames exceeds about 35.5°. Because the subjects' performance in some conditions exceeded expectations based upon constraints imposed by the structure-from-motion theorem, it is suggested that additional algorithms or heuristic rules might need to be considered when interpreting human recovery of structure in such displays. © 1987 Psychonomic Society, Inc.