How does noise influence the estimation of speed?

Local motion signals have to be combined in space and time, to yield a coherent motion percept as it is involved in a variety of visual tasks. This combination necessarily means to trade-off between loosing spatio-temporal resolution by pooling local signals and maintaining perceptually significant segmentation between them. When signals are pooled to detect the presence of coherent motion in large amounts of random noise, the question raised is how the noise affects the perceived quality, in particular speed, of the coherent motion. Is there an analogy to the well-known reduction in the perceived speed of moving gratings at low contrast? Using a two-interval forced-choice procedure, we have investigated the assessment of speed in random-dot kinematograms containing different proportions of noise. Under the conditions investigated, there is no strong reduction of perceived speed with increasing noise, as long as coherence levels remain well above the thresholds for directional judgements. This basic result, which could suggest considerable but not perfect segregation of signal and noise motion components in the pooling process leading to speed estimation, is discussed in relation to a model that is designed to decode speed from a population of elementary motion detectors (EMDs) of the correlation type. A strategy to estimate speed from a set of EMDs with a variety of spatio-temporal tuning does not only provide a velocity predictor unambiguous with the spatial structure of the stimulus, but also is largely independent of noise.