Abstract
One main goal of visual motion computation is to estimate the trajectory of objects moving in the scene from retinal optical flow. This is a tough computational problem under real-world conditions because retinal optical flow drastically changes with the optical material properties of the moving object. Specular and diffuse reflections, as well as refractions at object surfaces can produce complex patterns of optical flow that do not correspond with the object motions. In addition, these complex flow patterns vary with object shape and surrounding illumination. In this study we are investigating how constant we are in perceiving object motion across various contexts and if we compensate for other causal sources in motion. To see how much perceived object motion is independent of, or dependent on these factors, we asked twelve naïve observers to compare the rotational speed of two objects (T: test and M: match). M has matte shading with a texture, “knot” shape, and “forest” illumination map, while T was chosen from combinations of ten optical properties (e.g., matte, glossy, translucent), three shapes (knot, cubic, blobby), and three illumination maps (sunny, cloudy, indoor). The object rotation was around the vertical axis of the object at 0.5 rotation/sec for T, and variable for M. The exposure of each object was 500ms, and the PSE was estimated by a 2-IFC staircase method. We find illusory differences in perceived rotational speeds for different material, illumination, and shape conditions (e.g., transparent materials appear to rotate faster). Low-level optical flow models using horizontal optical flow and optical flow gradients explain 56% of the variance of the perceived speeds, but the magnitude of misperception predicted by the optic flow is much larger than the observed effect. These results suggest that the human visual system only partially compensates for effects of optical contexts in object motion.