A grouping process for motion segmentation

Purpose: Humans may use motion to segment the visual array into surfaces. Previous research on motion segmentation has focused on the detection of the velocity gradients that occur at the boundaries between surfaces. Here, we seek evidence for a complementary segmentation strategy that groups motions arising from the same surface. Methods: Moving dots were displayed in eight windows arranged in a circle. The speed of the dots in seven of the windows (the ground) was the same, while the speed of the dots in one window (the target) differed by a variable amount. The subject's task was to locate the target window. On each trial the direction of the dots in one window was chosen randomly. The directions of the dots in the remaining windows were then assigned using one of five rules: 1. direction was the same in all windows (translation) 2. direction decreased by 45 deg steps, moving clockwise around the display (spiral) 3. direction increased by 45 deg steps, moving clockwise (shear) 4. direction changed by 45 deg with both increasing and decreasing steps 5. direction varied randomly across windows Note that the ground in conditions 1-3 produces a global affine motion, but only conditions 1 and 2 are consistent with a rigid surface. Results: Subjects required a significantly smaller speed difference to locate the target in conditions 1 and 2 than in conditions 3-5. Conclusion: In condition 1, the velocity gradient between neighboring windows is largest around the target window, and so subjects may locate the target by locating the largest velocity gradient. If subjects used this strategy for all conditions, performance should have been the same for conditions 2-4, since the velocity gradients in these stimuli were identical. The superior performance for condition 2 suggests the existence of an alternative segmentation process that groups motions that are consistent with a single rigid plane.