Abstract
Cue integration has been demonstrated to be close to optimal under temporally constant stimulus conditions. That is, cues are assigned different weights according to their relative reliabilities. In real-world situations, however, stimulus conditions constantly change. For example, depending on the viewing situation the reliability of cues may change over time. Here we ask whether the system takes such continuous changes in reliability into account by adjusting the cue weights online. Subjects were binocularly presented with a spinning disk slanted in depth. Thus one cue was disparity, the other motion. There was a ±30 deg conflict between the slants defined by the two cues. We varied the reliability of the motion cue by sinusoidally changing the speed of rotation at different frequencies (0.067, 0.1, 0.2 Hz). Decreasing the speed of rotation decreases the reliability of the motion cue. However, it does not affect the magnitude of slant specified by the motion cue. Subjects task was to continuously adjust the angle of a two-lines probe according to the perceived slant. We found that increasing the motion cue reliability with faster rotations biased perceived slant towards the slant defined by the motion cue. The surface was therefore perceived to oscillate in depth according to the modulation of speed. The oscillation amplitude decreased with higher modulation frequency. The phase shift between rotation modulation and perceived oscillation increased with frequency. As a control, we repeated the task in order to estimate subject's reaction time for adjusting the probe. In this control the slant of the surface was actually oscillating in depth. By subtracting the reaction time from the phase shifts obtained in the experimental conditions we estimated that the time it takes to update the weights is less then a second.
FP6 EC project ImmerSence (IST-2006-02714)