Abstract
One issue that has largely been ignored in the cue combination literature is that different visual cues are processed with different latencies. Differences in latency could lead to errors in cue combination if the attribute that is being judged changes over time. Are differences in latency taken into account when combining different cues, or are errors arising from combining information from different moments simply tolerated? Does the brain reduce such errors by integrating cues over extended periods of time?
In order to answer these questions, we examined how varying the relative timing of changes in two slant cues influences the way that people perceive changes in slant. We presented subjects with a ring that jittered in a certain plane. The slant of this plane was evident from both binocular disparity and from the ring's retinal shape. The plane could change its slant as indicated by only one cue, by both cues simultaneously or by both cues at different times. In separate experiments subjects had to perform a detection task (detect the occurrence of a change in slant) and a discrimination task (indicate the direction of a change in slant).
In the detection task, the performance when both cues changed matched the performance that is predicted by combining the performance for the separate cues on the basis of probability summation, regardless of the timing of asynchrony. In the discrimination task, performance was significantly better than predicted by probability summation as long as the two cues changed within about 200 ms of one another. These findings suggest that rather than compensating for delays, the brain integrates cue estimates over an extended period of time. Our findings also show that cue integration is not completely compulsory; the benefit that we found for bi-cue slant changes in slant discrimination was absent in the detection task.