Effects of visual cues on flow parsing and simultaneous heading perception

Flow parsing, the ability to identify scene-relative object motion during self-motion, and heading perception are both essential components of locomotion. The causal inference hypothesis proposes that when observers cannot accurately identify object motion, they pool it with the background optic flow for heading estimation, leading to increased errors. Conversely, when they can correctly identify object motion, they segregate it from optic flow for heading estimation, leading to decreased errors. Here we tested this hypothesis and examined the effects of different visual cues on flow parsing and simultaneous heading estimation. On each 1-s trial, the display (90°×90°) simulated an observer translating toward a 3D cloud (0.56-5 m in depth) composed of 162 randomly positioned dots with an opaque window (16°×16°) containing 5 object dots at 8° or 16° offset from the simulated heading. These object dots moved laterally (0-0.8 m/s, lateral condition) at a fixed distance from or simultaneously approached (approaching condition) the observer. At the end of the trial, participants first reported whether the object dots moved left or right in the scene and then their perceived heading direction. Reducing simulated self-motion speed, increasing object lateral speed or dot density, decreasing the object position offset from the simulated heading, and incorporating binocular disparity cues all improved flow parsing performance (both accuracy and precision) but did not necessarily improve heading estimation. A significant correlation between the highest uncertainty in flow parsing and the largest heading errors supporting the causal inference hypothesis was observed only when the object dots underwent lateral motion while simultaneously approaching the observer. The results show differential effects of visual cues on flow parsing and heading perception, suggesting separate processing mechanisms for the two tasks. The causal inference hypothesis for flow parsing and heading perception can explain the data for some but not all types of object motion.