Self-motion cues in the natural habitats of zebrafish support lower visual field bias

Abstract

The optomotor response (OMR) behavior in larval zebrafish comprises swimming prompted by visual motion cues. This behavior is likely key to self-stabilization in moving aquatic environments. Larval zebrafish collect visual information from a very large field of view (>120 degrees both horizontally and vertically), but they preferentially respond to motion in the lower and posterior visual field with OMR (Wang et al., Cell Reports, 2020). We hypothesized that this preference reflects an adaptation to gather sensory signals from regions in the visual field where optical flow is most reliably informative about self-motion. To test this hypothesis, a video dataset was collected in nine natural shallow-water habitats located across the ecological range of the species. The videos were recorded using a 360-degree underwater camera, attached to a robotic gantry that moved through a series of controlled trajectories, including rotations and translations at multiple speeds. The total dataset includes several minutes of data for each trajectory, collected at 100 frames per second. Using an optical flow analysis on this dataset, we mapped the spatial distribution of several metrics of motion cue quality. We show that the lower visual field contains more texture information, produces more accurate motion cues over the dataset, and provides lower variance estimates of image motion from frame to frame. These factors likely contribute to more accurate and reliable self-motion estimates from the lower visual field compared to the upper visual field. The motion statistics found in our dataset both explain the lower-field bias seen in zebrafish OMR behavior and suggest practical design strategies for visual guidance systems in underwater robotics applications.