Depth Perception: Understanding Three-Dimensional Space through Visual Cues, Lecture notes of Earth Sciences

Download Depth Perception: Understanding Three-Dimensional Space through Visual Cues and more Lecture notes Earth Sciences in PDF only on Docsity!

Depth Perception

Perception & Attention Course

George

Mather

Depth cues

Retinal images are two

dimensional, yet the world is three

dimensional. The visual system must recover the missing dimension.

A depth cue can be defined as anything that is used by the visual system to estimate the depth of a point in space, or to perceive depth in a 3

D shape.

Retinal images contain multiple visual cues.

Oculomotor

cues are also available.

Visual cues can be divided into monocular and binocular.

Linear Perspective

Parallel lines sloping away from the observer converge as they recede into the distance.

Similarly sized shapes decrease in retinal size as they recede into the distance.

Blur

Eyes (and cameras) have limited depth of field

objects nearer or farther than the point of fixation or focus are blurred.

Blur variation across the image gives a cue to distance.

Accommodation

As fixation shifts from far to near, the shape of the lens must be changed to maintain a sharply focused image on the retina (accommodation).

Ciliary muscles are used to control lens shape. - Information on the state of the muscles offers a cue to fixation distance.

FAR

NEAR

Motion Parallax

As the observer moves, points at different distances move at different velocities over the retina.

Points beyond fixation (red) move in the direction of observer motion; points nearer than fixation (green) move in the direction opposite to observer motion. F T im e 1 T im e 2 F ie ld o f v ie w

Interposition

Near objects often partially obscure far objects.

The retinal image contains ‘T

junctions’ at points where contours of the nearer object intersect contours of the far object.

Convergence

Convergence angle is the angle formed by the two eyes when fixat ing an object at a specific distance.

As fixation shifts from a far object to a near object, convergen ce angle increases. - Extra

ocular muscles control eye position, including convergence angle

Information on the state of the muscles offers a cue to fixation distance. F a r F ix a tio n Ne a r F ix a tio n F F

Disparity in detail •^

Images of points at the same distance fall on corresponding retinal positions in the two eyes.

The horopter is a line drawn through all such points. - Images of points nearer or farther than fixation fall on non

corresponding or disparate retinal locations.

The sign or direction of the disparity signifies near vs. far depth. - The magnitude of the disparity signifies magnitude of depth. Ho r o p te r L e ft Re tin a Rig h t Re tin a F o v e a F F a r th e r a w a y : Un c r o s s e d o r fa r d is p a r ity F ix a te d : C o r r e s p o n d in g p o in ts o r z e r o d is p a r ity Ne a r e r : C r o s s e d o r n e a r d is p a r ity L e ft Re tin a Rig h t Re tin a F o v e a F

Stereoscopes

A variety of devices simulate the slightly different views creat ed by a real stereo image. Na tu r a l S te r e o Im a g e P r is m S te r e o s c o p e Re d -G r e e n An a g ly p h L e ft E y e V ie w Rig h t E y e V ie w

Disparity coding in the cortex

Primate cortex contains binocular cells that respond selectively to stimuli falling on disparate retinal locations in the two eyes.

Position of right

eye receptive field (RED) relative to left

eye receptive field (GREEN) determines preferred disparity.

Data shown are based on Poggio & Talbot (1981)

J.

Physiol

.^

**Stimulus C ell R esponse L R

deg (n e a r ) 0 deg -0. deg (fa r ) D isparity C ell 1 C ell 2**

Psychophysical evidence for disparity coding cells

Random dot stereograms (RDS) contain only disparity cues.

A sub

set of dots in one eye’s view is displaced relative to the other eye’s view.

Subjects perceive the displaced dots as standing out in depth against the other dots. - The success of RDS is evidence for a pure disparity coding mechanism.

Cue combination

Natural images contain multiple depth cues. How are they combine d?

The magnitude of different cues is very highly correlated: cue magnitude is proportional to (1/distance). - To derive a single estimate of depth from multiple cues, the vis ual system seems to take the average of different cue values. - Cues are weighted according to their reliability in a given set of stimulus conditions.