Spatiotemporal integration in the detection of coherent motion

We present data on the signal to noise threshold (SNR) for the impression of coherent, smooth motion in stroboscopically moving dot patterns (frame repetitionrate 100 Hz) degraded with additive noise (uncorrelated pattern each new frame). (Both patterns were 250 X 255 square arrays in which a pixel was randomly assigned a light or dark value with probability 50%.) It is found that the threshold varies roughly as the inverse square root of the number of available basic dot correlation pairs, subject to certain constraints that can be interpreted as due to the geometry of basic correlator units subserving the detection of motion. This geometry is estimated from the data and compared with values that have been obtained with independent methods. Phenomenologically the results are simple: when the target field is enlarged in the direction perpendicular to the motion the SNR threshold follows the theoretical expectations. Otherwise the variation is stronger, which points to a recruitment of new units. For an impression of movement to occur the smallest targets are short thin strips elongated in the direction of motion (e.g. 2.5' X 12' for 2.08 deg sec-1), but the threshold continues to fall at least to 5 degrees X 5 degrees. There is a trade-off between target area and presentation time. For a 5 degrees X 5 degrees target coherent motion can be seen in 3 successive frames, but for a 10' X 10' target ten times as many frames have to be presented, whereas an impression of movement is not obtained for still smaller targets.