Brain Motion Detection: Reichardt Detectors to Bayesian Inference

The visual system detects motion with direction‑selective detectors that compare spatially displaced inputs after precise time delays, a principle captured by the Reichardt model. V1 neurons act as orientation‑selective filters in space‑time, encoding one‑dimensional velocity components. Area MT receives these 1D signals and combines them to represent two‑dimensional pattern motion, a distinction revealed by plaid‑stimulus experiments that separate component‑selective (V1) from pattern‑selective (MT) responses.

The Aperture Problem

Local measurements of an edge are geometrically ambiguous: many velocity vectors share the same perpendicular component, creating the aperture problem. The visual system resolves this ambiguity by integrating motion information across space and by exploiting unambiguous two‑dimensional features such as corners, occlusion boundaries, and contour continuity. Integration is not automatic; it is gated by perceptual organization that signals the presence of occlusion or relatable contours.

Bayesian Motion Perception

Perceptual inference follows a Bayesian rule: the posterior probability of a velocity equals the product of a sensory likelihood and an internal prior. A prior that favors slow speeds explains why low‑contrast stimuli appear to move slower than high‑contrast stimuli. When contrast is low, the likelihood becomes broad and fuzzy, allowing the slow‑speed prior to dominate the posterior and bias the perceived velocity toward zero.

Motion, Depth, and Motor Control

The brain interprets motion through a rigidity prior, often inferring three‑dimensional structure from two‑dimensional motion cues such as the kinetic depth effect. Optic flow patterns are used to calibrate posture; moving walls can shift adult posture and cause falls in children. To avoid mistaking retinal motion caused by eye movements for external motion, the brain discounts it using efference copies of motor commands rather than relying on proprioceptive feedback.

Mechanisms & Explanations

A Reichardt detector receives spatially offset inputs that are delayed to match the temporal offset of a moving stimulus; simultaneous arrival triggers a direction‑selective response. The intersection‑of‑constraints method resolves two‑dimensional motion by finding the common point where multiple one‑dimensional constraint lines intersect. Bayesian motion inference multiplies a contrast‑dependent likelihood by a slow‑speed prior; high contrast yields a sharp likelihood peak, while low contrast produces a broad likelihood that lets the prior shift the perceived velocity toward slower values.