Abstract
Many organisms and objects deform when moving, requiring perceivers to separate shape changes from object motion. We have shown that 3-D shape extraction from motion cues is as good for non-rigid as for rigid objects. Here we address observer abilities to identify different classes of dynamic shape deformations. Point-light cylinders that were either rigid (with assorted axial curvatures) or flexing in depth or in the image plane, were rotated simultaneously in depth and the image plane, and presented monocularly in perspective projection. Texture and density cues were removed by placing point-lights randomly on the 3-D surface after rendering. Using a method of constant stimuli, the amount of flex was varied across trials, and observers had to classify each cylinder as rigid, flexing in depth, or flexing in the image plane. Due to the complex rotation and the curved shapes, the projected contours of cylinders of all three classes varied in curvature during the trial, so contour deformation was not informative for identifying the class. Results were consistent across three observers. Depth-flex cylinders were perceived as rigid for low values of non-rigidity and as depth-flex for higher values. Plane-flex cylinders were confused with depth-flex cylinders for low non-rigidities and were perceived veridically for high values. Rigid cylinders bent in depth were slightly confused with depth-flex cylinders, but rigid cylinders bent in the plane were seen as rigid. There are no published models that identify shape deformations from motion cues. Eigen-shapes extracted from image sequences can distinguish between rigid and flexing cylinders, but not the viewer-centered distinction between depth- versus plane-flex. We combine relative image velocities into the differential invariants div, curl, and def. We show that the gradient of def is zero for rigid but non-zero for flexing cylinders, and explore combinations with curl and div that could classify dynamic deformations.
NEI grants EY013312 & EY07556.