Abstract
When an oriented line moves in depth inside a circular aperture motion direction of this line is highly ambiguous. From a computational point of view both motion and stereo correspondence is under-determined yet somehow the visual system solves the inverse problem. Although observers reliably report azimuth and elevation angle of 3D motion direction (Heron & Lages, VSS 2009) existing models can only predict azimuth of horizontal motion in depth and face computational problems when trying to explain non-horizontal motion in depth.
The interocular velocity difference (IOVD) model detects local motion vectors in the left and right eye before it computes motion direction. It is shown that a standard IOVD model offers no general solution to the inverse problem of an oriented line moving in depth. If the model uses velocity constraints rather than motion vectors additional constraints, such as minimal displacement in 3D, have to be introduced to establish a 3D motion direction.
Changing disparity over time (CDOT) on the other hand suggests that horizontal disparities are extracted to build a depth map of the line stimulus that can be tracked over time. It is shown that the average of left and right eye motion vectors combined with corresponding horizontal disparity provides a suitable estimate of 3D motion direction in a binocular viewing geometry.
We compared predictions of a generalized IOVD and CDOT model with empirical data from two experiments. It is concluded that both generalized models still have difficulties to predict perceived motion direction of an oriented line. Additional geometric constraints are discussed that may be able to explain binocular 3D motion perception.