Abstract
Purpose: In a heading perception task, we demonstrated that although error in the perceived heading direction is modulated by the range of directionally constrained noise, humans are proficient in integrating the noisy flow information to form a global percept. (VSS2005) To investigate how much of the information available in the velocity field is useful to observers, we now propose two ideal observer models (IOM).
Experiment: In a display of expanding random-dot-kinematograms (RDK) perturbed with directionally constrained noise, observers are required to determine the direction of straight-line-heading (left or right of a vertical reference line appearing after termination of motion). Accuracy of heading perception is measured by an adaptive-staircase.
Model: Given the vertical line's position and distance to the heading direction, the IOM decides whether the heading is to the right or to the left of the vertical line. We constructed a 3D-IOM which has 3D information about every dot location and a 2D-IOM which uses the projections of dot locations onto the computer-screen. The IOMs are applied to two experimental conditions we previously used in psychophysics: 1)Random-walk: Dot directions are independently perturbed frame-to-frame; 2)Fixed-trajectory: Dots keep same perturbed direction throughout their lifetimes.
Results: a) Fixed-trajectory condition is harder for both the human observers and IOMs. b) Heading perception accuracy in 3D-IOM is robust to increases in direction range, while in 2D-IOM it rapidly drops. Observers' efficiencies from 2D-IOM are higher than from 3D-IOM.
Conclusion: 2D-IOM results better resemble human data, thus reconstruction of 3D-environment is not necessary for heading perception
NIH grant R01EY007861-15 L.M.V.