Processing of sensory signals by the brain is subject to physical limitations, both external (statistical quality of sensory input) and internal (physiological limitations on neural signals). Efficient use of limited resources requires that the brain match its processing strategies to the statistical structure of input signals. We test this idea in the context of motion estimation in the fly visual system, where the optics of the compound eye and signals and noise in the retina are well understood and we can record from output neurons that encode velocity estimates. Still missing is a full characterization of the statistical relation between visual signals and motions relevant for flies in a natural context. Therefore, we develop a specialized camera mimicking fly eye optics, with inertial motion sensors providing ground truth about motions. We describe the design, construction, and performance characteristics of this FlEye camera. From camera data sampled in nature we construct optimal local motion estimators. These estimators show characteristic biases that are also observed in flies and other biological systems. Physical limitations of the camera data and computational estimator are negligible compared to biological systems. That we nevertheless observe similar biases suggests that biological performance is effectively limited by external statistics, not physiology.

Close

PrevNext

View figure in article

• Biological information processing
• Motion detection
• Neuroscience, neural computation & artificial intelligence
• Sensory processes
• Vision