Perception of nonrigid structures from motion using tracking image gradient vectors.

Introduction: A subset of the true optical flow can be extracted by constructing a vector field that represents image gradients and then tracking vectors in this vector field. This pseudo-flow (p-flow) subset effectively visualizes nonrigid motion and leads to the perception of nonrigid structure from motion. In this study, we investigate whether the human sensory system can extract information about the physical properties of inanimate fluid, especially viscosity, from the p-flow. Methods: Computer-generated movies of flowing liquid were constructed using the p-flow algorithm and the Lucas-Kanade method. The movies featured liquids of different viscosities in the form of point-light displays. The viscosity of the fluid in various subsets of these movies was then estimated by 312 participants. Results: The error, i.e., difference between expected and actual ratings showed smaller variability across repeated trials and the mean response time was significantly shorter when using the p-flow than with the conventional Lucas-Kanade method. Discussion: Our results suggest that the p-flow enables a more reliable viscosity rating, which could be related to the local constraint used in the algorithm.