Storyboard

The detection and estimation of motion own and external objects is one of the most fundamental computations of the visual system. The optical flow (Horn-Schunck) equation states the basic constraint: the intensity of a point does not change between frames (I_x·u + I_y·v + I_t = 0), but this is ambiguous for a single edge orientation (aperture problem).

The Lucas-Kanade estimator resolves the ambiguity by accumulating the optical flow constraint over a spatial window and solving by least squares. The result is robust when the image has gradients in multiple directions (corners, textures). The vertebrate visual system implements similar computation in the V1MTMST hierarchy.

The time to impact _TTC = / is a particularly useful optical invariant for landing control, object avoidance, and prey capture. It does not require knowing either the distance d or the velocity d/dt separately, only the angular expansion directly observable on the retina. Bees use _TTC to regulate landing speed; Falcons and lobsters use it to capture prey.

Reichardt's model is the motion detector implemented by insect visual systems (fly H1 neurons). Compares the signal from one receiver to the delayed signal from a neighboring receiver: the product detects spatio-temporal correlation indicating motion in one direction. It is directionally selective and free of contrast ambiguities.

Proportional navigation (PN) is the guidance law that maximizes the probability of intercepting moving prey. Adjusts lateral acceleration in proportion to the line of sight rotation rate (d/dt): if the line of sight rotates, the prey does not converge toward the intercept and correction is needed. Dragonflies and hawks implement PN with N 35, very close to the theoretical optimum.

>Model

ID:('ky', 583)

Description

ID:(0, 583)