Purchase this article with an account.
Bo Hu, David Knill; Kinesthetic feedback helps disambiguate 3D structure-from-motion. Journal of Vision 2007;7(9):89. doi: https://doi.org/10.1167/7.9.89.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Visual motion provides information about the 3D shapes of surfaces, but is confounded with information about the 3D movement of surfaces. We tested whether the brain uses kinesthetic information about object motion produced when actively manipulating objects to disambiguate 3D structure-from-motion. Five naïve subjects monocularly viewed computer-rendered folded cards through a 4-degree aperture. The virtual surfaces were covered with random dots and shown in a mirrored view of a computer monitor, making the virtual surface appear behind the mirror. In active conditions, subjects rotated the virtual surfaces themselves by rotating a rod mounted behind the mirror in the same location as the virtual surface. The rotation was fed back into the computer to generate the rotation of the surface. In passive conditions, subjects were shown movies of the rotating cards sampled from the recorded rotation sequences from their active sessions. Subjects viewed surfaces with different dihedral angles and were asked to judge whether the angle was greater or less than 90 degrees. We built a Bayesian model for estimating dihedral angle from visual motion information that was parameterized by known noise parameters on sensed velocity and acceleration and was supplemented by noisy kinesthetic feedback when available. Under passive viewing, the model has a large bias toward acute angles. The bias is reduced by kinesthetic feedback from active rotation. Kinesthetic feedback also significantly reduces the variance of the estimator. Consistent with these predictions, subjects showed significantly lower biases toward acute angles in the active conditions than in the passive conditions (by ∼24 degrees). The std. deviations of their estimates, indicated by the slopes of their psychometric functions, were also significantly lower in active conditions than in passive conditions (average: 17 degrees vs. 28 degrees). The results provide strong evidence that humans use kinesthetic information to disambiguate 3D structure from visual motion.
This PDF is available to Subscribers Only