The Perception of Motion and Structure in Structure-from-motion: Comparisons of Affine and Euclidean Formulations

Abstract

I investigated the discrimination of rigid from nonrigid structure and the perception of affine stretches along the line of sight [Norman & Todd (1993). Perception and Psychophysics, 53, pp. 279-291]. Investigations of performance at discriminating rigid from nonrigid structure showed that performance improved when number of views and amount of simulated three-dimensional nonrigidity increased. Investigations of rotations about the vertical which include affine stretches along the line of sight compared Euclidean interpretations of affine-stretching stimuli to human perception. These Euclidean interpretations were obtained from a simple algorithm which recovered structure and motion from this limited class of stimuli under the assumption that distances to the axis of rotation did not change. The algorithm predicted that stretches along the line of sight would be perceived as nearly rigid and have variable angular velocity. These predictions were supported by subjects' reports of occurrences of nonrigidity and minima of angular velocity. The Euclidean algorithm also provided measures of nonrigidity and motion coherence, and experimental results were consistent with a prediction of when perception of nonrigidity would be independent of perception of coherence. The results are discussed relative to the advantages and shortcomings of both the affine and Euclidean approaches to structure-from-motion.

Hearing a face: Cross-modal speaker matching using isolated visible speech

“…One possibility is that the observers were recognizing visible speakers by extracting general facialstructure-from-motion information (e.g., Braunstein, Hoffman, & Pollick, 1990;Christie & Bruce, 1998;Hildreth, Grzywacz, Adelson, & Inada, 1990;Pollick, 1997) and then matching it to the heard voice that best fit that facial structure. Although the results reported above are consistent with this interpretation, a post hoc analysis revealed results less consistent with a structure-from-motion basis.…”

Structure from motion: A tolerance analysis

“…The depth calculated from only three successive frames is very sensitive to noise in situations in which the vector is parallel to the viewing direction. Pollick (1993) also analyzed the stimuli used by Norman and Todd (1993). He assumed that the total displacement of the projection of a point is twice the distance (r) to the axis of rotation.…”

“…This algorithm works quite well in this situation. Pollick (1993) investigated more points and found only 1% of nonrigidity (measured as the variances of the interpoint distances). Pollick's (1993) algorithm appears to be rather ad hoc.…”

“…Pollick (1993) investigated more points and found only 1% of nonrigidity (measured as the variances of the interpoint distances). Pollick's (1993) algorithm appears to be rather ad hoc. A structure is deduced by assuming constant distance from the 3-D points to the axis of rotation.…”

Sol: An Agent-Based Framework for Cyber Situation Awareness

“…Instead of relying on the slow and small human focal vision system, OZ is designed to use the fast and robust ambient vision systemthe same system that people use to quickly and successfully navigate crowded hallways without conscious thought or to catch a football on the run [31,32,45,46]. As another example, OZ exploits the capabilities of human vision for quickly perceiving changes by using movement to convey difficult, correlated information [33,38].…”