Abstract
Liquids with different viscosities, such as milk, honey and hair-gel, respond to forces in radically different ways. Low viscosity fluids, like water, tend to flow and splash easily, whereas high viscosity fluids, like melting glass, tend to ooze slowly into distinctive clumps. These tendencies lead to strikingly different visual appearances, and in everyday life we readily distinguish between different materials based on the way they flow and change shape over time. Despite this, very little is known about how the brain estimates and represents the properties of fluids. Here, we use computer simulations of liquids with different viscosities to study the visual estimation of viscosity. This allows us to parametrically vary the material properties of the fluid while holding constant all other aspects of the scene (the fluid’s volume, initial position, velocity, and optical surface properties), and allows us to make detailed measurements of the fluid's behaviour to correlate with perception. The simulated fluid flowed out of a pipe and fell into a rectangular container. In addition to the viscosity, we also varied the height of the pipe (which affects momentum at initial impact) and the presence of a spherical obstacle in the fluid’s path. Using rating scales and maximum likelihood difference scaling (MLDS) we measured a 7-point psychophysical function relating the physical viscosity coefficient (across 7 log units) to perceived viscosity. We find the function is surprisingly close to linear and that subjects already have a strong impression of the viscosity only a few frames after initial impact. We compared performance to a number of 2D and 3D measurements derived from both optic flow and surface geometry. Together these analyses suggest a number of simple heuristics that the visual system could use to estimate viscosity from the way fluids move, and settle into specific shapes.
Meeting abstract presented at VSS 2012