Abstract

Structure from motion often refers to the computation of inverse optics from a matched sequence of images. However, as is suggested by human vision, a depth map of a surface is probably not a good representation for storage and recognition, a more concise representation seems necessary. In addition, recovering the exact function of the surface and the motion parameters is computationally expensive, serial, and sensitive to errors. Obviating this computation, I first show that the sign of the normal curvature in a given direction on a given point in the image can be computed from a simple difference of slopes of line segments in one image. This simple operator is an example of a hyperacuity task which the visual system can perform with higher, precision, an order of magnitude better than visual acuity. Using this result, local surface patches can be classified as convex, concave, cylindrical, saddle point, or planar. This is done by repeating the basic sign operator at different orientations, a natural operation to many known biological architectures. At the same time the translational component of the optical flow is obtained, from which the focus of expansion can be computed (which is important for navigation, for example).

© 1989 Optical Society of America